Acceptance Sampling & AQL Calculator

What acceptance sampling does

Acceptance sampling is the discipline of deciding whether to accept or reject a whole lot of product by inspecting only a random sample of it. Rather than check every unit, you inspect a defined number, count the defectives, and accept the lot if that count stays at or below an acceptance number, rejecting it otherwise. The appeal is economic: full inspection of a large lot is slow and expensive, and impossible when the test destroys the unit, so sampling trades a small, precisely quantified risk of a wrong decision for a large reduction in inspection effort. The art is choosing a plan whose risks are acceptable, and that is exactly what a standard like ANSI/ASQ Z1.4 provides.

The plan has three numbers: the sample size, the acceptance number, and the rejection number. Inspect that many units, and the decision rule is simple and absolute, accept if defectives are at or below the acceptance number, reject if they reach the rejection number, with nothing in between for single sampling. What makes the plan trustworthy is that its risks are known. Every plan has an operating characteristic curve that says, for any true lot quality, how likely the plan is to accept it, so you can see before you inspect a single unit how well the plan protects both the producer, who does not want good lots rejected, and the consumer, who does not want bad lots accepted.

This calculator implements ANSI/ASQ Z1.4 single sampling under normal inspection, the most widely used acceptance-sampling system, derived from the old MIL-STD-105 and mirrored internationally by ISO 2859-1 and Brazil’s NBR 5426. Enter the lot size, pick an inspection level and an acceptable quality limit, and it returns the sample size code letter, the sample size, and the acceptance and rejection numbers, then draws the OC curve and reports the producer risk. Give it a rejectable quality level and it also reports the chance a lot that bad would slip through. It handles the standard’s arrow rules automatically, so the plan it gives is the one the printed tables would give.

How this calculator works, step by step

Start with the lot size, the total number of units in the batch you are deciding on. Then choose an inspection level. General level II is the standard and the right choice unless you have a specific reason to change: level I gives a smaller sample and less discrimination when inspection is costly or less protection is acceptable, and level III gives a larger sample and more discrimination when more protection is worth the extra inspection. The lot size and level together determine the sample size code letter through the standard’s Table I.

Next pick the acceptable quality limit, the AQL, as a percent defective. This is the quality level the plan is designed to accept most of the time, and it should be set to the worst average quality you are willing to routinely pass, not to a target you hope to hit.

The calculator combines the code letter and the AQL to read the sample size, the acceptance number Ac, and the rejection number Re from the master table, resolving the up and down arrows the standard uses when a plan would fall off the table.

Optionally, enter a rejectable quality level, a higher percent defective you want to be sure of catching, and the tool computes the chance the plan would accept a lot that bad, the consumer risk.

The result panel headlines the sample size, then lists the code letter, sample size, acceptance and rejection numbers, the producer risk at the AQL, and the acceptance probability at your rejectable quality level. A plain-language rule states the decision directly: inspect this many units, accept on this many defectives or fewer, reject otherwise. The chart draws the operating characteristic curve, the probability of acceptance across the full range of true lot quality, so the plan behavior is visible at a glance. Download a PDF or CSV or share the plan; everything runs in your browser and nothing you enter is stored.

The standard behind the plan

ANSI/ASQ Z1.4 works in two lookups. First, Table I converts the lot size and inspection level into a sample size code letter, a single letter from A to R, where larger lots and higher levels give later letters and thus larger samples. The code letter deliberately separates the sample size decision, which depends on lot size and how much discrimination you want, from the AQL decision, which sets how strict the accept criterion is. Second, the master table for single sampling under normal inspection takes the code letter and the AQL and returns the sample size, the acceptance number, and the rejection number, which is always the acceptance number plus one.

The one subtlety is the arrows. When a code letter and AQL combination would call for a plan that does not exist sensibly, for instance a sample larger than the tabulated range or an acceptance number that cannot discriminate, the table shows an arrow pointing to the first workable plan above or below, and you use that plan, including its different sample size.

This is why the sample size the standard gives sometimes differs from the base sample size of the code letter. This calculator resolves those arrows exactly as the standard directs, so the plan it returns matches the printed tables without your having to trace the arrows by hand.

The acceptance decision itself is then a single comparison: count defectives in the sample, accept if the count is at or below Ac, reject at Re.

Five worked examples you can follow

Example 1: a standard mid-size lot

A lot of 5,000 units at general inspection level II and an AQL of 1.0 percent gives sample size code letter L, a sample of 200 units, with an acceptance number of 5 and a rejection number of 6. The rule is to inspect 200 randomly chosen units, accept the lot if 5 or fewer are defective, and reject it if 6 or more are. At the AQL of 1 percent the producer risk, the chance of rejecting a lot that is actually at acceptable quality, is only about 1.6 percent, so good lots almost always pass.

Example 2: a small lot

A lot of 100 units at level II and an AQL of 2.5 percent gives code letter F, a sample of 20, with an acceptance number of 1 and a rejection number of 2. Inspect 20 units, accept on 0 or 1 defective, reject on 2. The small sample means the plan discriminates weakly: its OC curve is shallow, so a moderately bad lot has a fair chance of acceptance. That is the price of inspecting few units, and it is why small lots carry more risk per decision than large ones.

Example 3: tightening the AQL

Take the same 5,000-unit lot at level II but tighten the AQL to 0.10 percent. Now the code letter L cannot support so strict an AQL with its 200-unit sample, so the standard’s arrow sends you to a larger plan: code letter M, a sample of 315, with an acceptance number of 0 and a rejection number of 1. A tighter AQL forces a larger sample and a stricter accept criterion, here accept only if the entire sample of 315 is defect-free, which is exactly how the standard delivers more protection when the quality bar rises.

Example 4: reading the consumer risk

Return to the first plan, n 200 with Ac 5, and suppose you consider 4 percent defective an unacceptable, rejectable quality level. Feeding that in, the plan accepts a 4-percent lot about 19 percent of the time, a non-trivial consumer risk showing that this AQL-1-percent plan does not strongly screen out 4-percent quality. Driving that acceptance probability down toward zero would require a larger sample, which is the trade the OC curve makes visible: bigger samples push the acceptance probability of bad lots lower, so if 4 percent must almost always be caught, either the sample or the AQL has to change.

Example 5: choosing an inspection level

Keep the 5,000-unit lot and AQL of 1.0 percent but switch to inspection level I. The code letter drops to J, a sample of 80, with an acceptance number of 2 and a rejection number of 3. The smaller sample costs less to inspect but discriminates less sharply than the level II plan of 200. Switching to level III instead would raise the sample toward 315 for finer discrimination. The level is the lever for trading inspection cost against the sharpness of the accept or reject decision, independent of the AQL.

Three expert tips for acceptance sampling

AQL: the most misunderstood number in sampling

No term in acceptance sampling causes more confusion than the acceptable quality limit, and getting it right is essential to using a plan correctly. The AQL is defined as the maximum percent defective that, for the purpose of sampling inspection, is considered satisfactory as a process average. The key phrase is process average: the AQL describes the quality level that a plan will accept most of the time, roughly 95 percent, when the supplier’s ongoing average sits there. It is the boundary of routinely acceptable quality, the point on the OC curve where the producer risk is small by design.

The pervasive mistake is to treat the AQL as a target or an allowable defect rate, as if an AQL of 1 percent meant one percent defective is the goal or the permitted level for any single lot. It is neither. A supplier who actually runs at the AQL will have a meaningful fraction of lots rejected and is skating at the edge of acceptability; a well-run process aims far below its AQL so that essentially every lot passes comfortably.

Nor does accepting a lot mean the lot is at or below the AQL, because acceptance is probabilistic: a lot somewhat worse than the AQL can still be accepted, just less often. The correct reading is that the AQL sets the strictness of the plan, and this calculator returns the Z1.4 plan matched to the AQL you choose, but the quality you should manage toward is better than the AQL, not equal to it.

Setting the AQL too loose to make lots pass simply guarantees more defects reach the customer.

Reading the operating characteristic curve

The operating characteristic curve is the single most informative thing about a sampling plan, because it shows the plan behavior across every possible level of incoming quality rather than at just one point. The horizontal axis is the lot true percent defective, and the vertical axis is the probability that the plan accepts the lot. Every plan produces the same general shape: high acceptance probability for very good lots on the left, falling through a sloping middle region, to near-zero acceptance for very bad lots on the right. Where and how steeply the curve falls is what distinguishes a strong plan from a weak one.

Two points on the curve carry the risks that matter. At the AQL, on the good-quality side, the curve should sit high, and the small gap below the top, typically around 5 percent, is the producer risk: the chance a lot of acceptable quality is rejected by bad luck of the draw. Farther right, at whatever defect level you consider genuinely unacceptable, the curve height is the consumer risk: the chance a bad lot is accepted anyway. A good plan keeps both small, which requires a steep curve, and a steep curve requires a large sample.

This is the fundamental trade the OC curve makes visible: for a fixed sample size you cannot lower both risks at once, and the only way to sharpen the plan’s discrimination, steepening the curve so it separates good from bad more decisively, is to inspect more units.

This calculator draws the OC curve for your exact plan and marks the producer risk at the AQL, so you can judge before inspecting whether the plan protects you adequately against the quality levels you actually worry about, and enter a rejectable quality level to read the consumer risk directly.

Producer and consumer risk, and the balance between them

Every sampling plan embodies a negotiation between two parties with opposite fears, and the two risks name those fears precisely. The producer, the supplier or the upstream process, fears that a lot which is genuinely good, at or better than the AQL, will be rejected by the sample through ordinary sampling variation, incurring cost and delay for no real quality reason.

That is the producer risk, and it is measured at the AQL where the plan is designed to keep it small, conventionally near 5 percent. The consumer, the customer or the downstream operation, fears the opposite: that a lot which is genuinely bad will be accepted because the sample happened to contain few defectives.

That is the consumer risk, measured at the higher defect level the consumer considers unacceptable, sometimes called the lot tolerance percent defective or rejectable quality level.

These risks are linked, and the link is the reason acceptance sampling is a balance rather than a free lunch. For a given sample size, tightening the plan to reduce the consumer risk, by lowering the acceptance number, raises the producer risk, because now some good lots also get rejected; loosening it does the reverse.

The only way to reduce both at once is to increase the sample size, which steepens the OC curve and separates the two quality levels more sharply, at the cost of more inspection. Z1.4 encodes a sensible default balance for each combination of lot size, level, and AQL, which is why using the standard plan is usually wiser than inventing your own.

This calculator makes the balance concrete by reporting the producer risk at your AQL and, when you supply a rejectable quality level, the acceptance probability there, so you can see both sides of the negotiation the plan represents and judge whether its particular balance fits your situation.

When to sample and when to inspect everything

Acceptance sampling is powerful but not universal, and knowing when it is the wrong tool is part of using it well. Sampling is the right choice when full inspection is impractical or wasteful: when the test destroys the unit, so inspecting everything would leave nothing to ship; when volumes are so high that checking every unit is prohibitively slow or expensive; or when the process is stable and the incoming quality is consistent enough that a random sample fairly represents the lot. In those cases a good plan gives strong protection for a small fraction of the inspection cost, which is precisely the economic case the method was built for.

There are situations where sampling is not appropriate, and two stand out. When a defect is critical, meaning a single escape could cause injury, a safety failure, or a severe cost, the residual consumer risk of any sampling plan may be unacceptable, and 100 percent inspection, or better, error-proofing the process so defects cannot occur, is warranted.

And when the process is out of control or its quality is erratic, the premise of sampling breaks down, because a random sample from a lot whose defect rate swings unpredictably does not reliably represent it.

It is worth remembering, though, that 100 percent inspection is not perfect: human inspectors miss defects through fatigue and monotony, so screening every unit typically catches only 80 to 90 percent of defects, and a well-designed sampling plan on a stable process can sometimes deliver better outgoing quality than tired eyes checking everything.

The decision is therefore not simply sample versus inspect all, but which approach, given the defect’s criticality, the process stability, and the realities of inspection, actually delivers the outgoing quality you need. This calculator sizes the sampling option; the judgment of whether to sample at all rests on those factors.

Switching rules and the bigger system

A single sampling plan is a snapshot, but Z1.4 is designed to run on a continuing stream of lots from one supplier, and its switching rules are what give it teeth over time.

The standard defines three severities: normal inspection, which this calculator computes and which applies by default; tightened inspection, with stricter acceptance numbers, invoked when recent lots show quality slipping; and reduced inspection, with smaller samples, permitted when a supplier’s quality has been consistently good.

The rules move you between these automatically based on the recent history of accepted and rejected lots, so a supplier who lets quality drift is met with tighter scrutiny, and one who performs well earns lighter inspection.

This structure turns a set of static plans into a dynamic system that rewards good quality and penalizes bad, applying continuing pressure toward improvement rather than merely judging one lot at a time. It matters because the protection of any single plan is limited, but the switching rules add a second layer: a supplier cannot ship marginal quality indefinitely without triggering tightened inspection, whose steeper acceptance criteria catch more bad lots.

The practical implication is that the normal-inspection plan this calculator provides is the starting point and the most common case, but in a mature incoming-inspection program it should sit inside the switching framework, with tightened and reduced plans held in reserve and applied per the rules.

Understanding this keeps the single plan in perspective: it is one state of a system whose real power is the way it adapts to a supplier’s demonstrated quality over a run of lots.

Single, double, and sequential sampling

The plan this calculator produces is a single sampling plan, the simplest and most common form, where one sample decides the lot. Z1.4 also defines double and multiple sampling plans, which can reduce the average amount of inspected product for the same protection, at the cost of more procedural complexity.

In double sampling a first, smaller sample can accept a clearly good lot or reject a clearly bad one outright, and only an ambiguous result in between triggers a second sample; the two samples together then decide.

Multiple and sequential sampling extend this idea to several stages, drawing units until the accumulated evidence is decisive, which minimizes the expected number inspected but requires careful administration.

The reason these alternatives exist is that they exploit easy decisions. Many lots are clearly good or clearly bad, and for those a small first sample suffices, so on average fewer units are inspected than a single fixed sample would require, even though the worst-case inspection can be larger. The trade is operational: double and multiple plans need trained inspectors and disciplined procedures to handle the staged decisions correctly, and the arithmetic is less transparent.

For most inspection decisions the single sampling plan is the right default, both because it is simplest to execute and audit and because its OC curve, the one this calculator draws, is straightforward to interpret.

Where inspection cost is high and volumes justify the added complexity, double or sequential sampling can lower the average inspection burden, and the same standard provides those plans; but the single-sampling plan here is the workhorse of acceptance sampling and the correct place to start.

Common mistakes in acceptance sampling

A handful of errors recur and undermine sampling decisions. Watch for these.

• Treating the AQL as a target. The AQL is the worst routinely acceptable average, not a goal. Manage the process to well below it.
• Non-random sampling. Pulling units from one spot, the top layer, the first off the line, breaks the statistics. Every unit must have an equal chance of selection.
• Ignoring the OC curve. The plan numbers are meaningless without the curve. Check the acceptance probability at quality levels worse than the AQL.
• Assuming acceptance means good. Accepting a lot does not prove it meets the AQL; a somewhat worse lot can still be accepted. Acceptance is probabilistic.
• Sampling critical defects. For defects that could injure or cause severe cost, the residual consumer risk may be unacceptable; consider 100 percent inspection or error-proofing.
• Wrong inspection level. Using level I to save inspection when more protection is needed, or level III when it is not, mismatches sample size to the decision. Level II is the default for a reason.
• Ignoring switching rules. Running normal inspection forever misses the system’s power. Apply tightened and reduced inspection per the supplier’s history.

Where acceptance sampling fits in the quality toolkit

Acceptance sampling sits at the boundary between organizations, or between process stages, where lots change hands and a keep-or-return decision must be made. It is a gatekeeping tool, not a process-improvement one: it decides the fate of product already made, whereas the other quality tools work to make that product right in the first place. This distinction matters, because the long-run answer to a supplier who keeps sending marginal lots is not more sampling but better process control at the source, and acceptance sampling is best seen as a safety net and a negotiating framework rather than a substitute for capability.

That said, it connects naturally to the rest of the toolkit. The defect rate a lot presents relates to the supplier’s process sigma, which the sigma level and DPMO calculator expresses, and to their first-pass yield, which the process yield calculator rolls up across steps.

Whether the supplier’s process is even stable enough for sampling to be meaningful is a question for the control chart calculator, and whether it can meet the specification is what the process capability calculator answers.

In a mature relationship, evidence of a capable, in-control process can justify reduced inspection or even skip-lot sampling, so the process tools and the sampling plan reinforce each other: good capability lightens the inspection burden, and sampling results feed back pressure to improve capability. Return to the Quality Control hub for the full set.

The origin and reach of Z1.4

The acceptance-sampling tables in Z1.4 descend from military necessity. During the Second World War the United States military faced the problem of accepting vast quantities of munitions and materiel without inspecting every item, and statisticians developed sampling procedures that became MIL-STD-105.

Refined through several revisions, that standard was later adopted by civilian standards bodies and became ANSI/ASQ Z1.4, while the internationally harmonized version is ISO 2859-1 and national equivalents exist worldwide, including Brazil’s NBR 5426.

The lineage explains the standard’s structure, the code letters, the switching rules, the emphasis on a continuing stream of lots, all of which were designed for a large buyer receiving ongoing deliveries from many suppliers.

Its durability comes from being a complete, self-consistent system rather than a single formula.

By fixing the relationship between lot size, sample size, and acceptance criteria across a wide range of AQLs and inspection levels, and by adding the switching rules that adapt to demonstrated quality, Z1.4 gives a buyer and a supplier a common, auditable language for the accept-or-reject decision.

That shared framework is why it remains the default for attribute acceptance sampling across manufacturing, distribution, and global trade decades after its military origin, and why the plan this calculator produces will match what an inspector reading the printed tables, or a supplier in another country using ISO 2859-1, would independently arrive at.

Drawing the sample: the step that voids the plan if done wrong

Every risk figure a sampling plan reports rests on one assumption that lives entirely outside the arithmetic: that the sample is genuinely random, meaning every unit in the lot had an equal chance of being chosen. The whole statistical apparatus, the OC curve, the producer and consumer risks, the binomial probabilities, is built on that premise, and when it fails the numbers become fiction while still looking authoritative. This makes sampling technique, not calculation, the most common real-world failure point of acceptance sampling.

The ways it fails are mundane and easy to fall into. Pulling the sample from the top layer of a pallet, the first units off a run, or the most accessible cartons gives a convenience sample, not a random one, and if defects cluster, by shift, by machine, by position in the batch, such a sample can systematically miss them and accept lots it should reject. A sample taken only from material the supplier presents for inspection is worse still, because it invites selection of the best units.

The remedy is a deliberate randomization method: number the units or containers and use random numbers, or sample across the full spatial and temporal spread of the lot, so no region is over- or under-represented. It is also why lot formation matters, a lot should be homogeneous, produced under uniform conditions, so that a random sample can fairly represent it; mixing product from different shifts or machines into one lot undermines both the randomness and the meaning of the result.

The calculator sizes the sample and sets the accept rule, but the protection it computes is only real if the units inspected were drawn at random from a properly formed lot.

Units and quick reference

Enter the lot size as a whole number of units and the AQL and rejectable quality level as percentages; the sample size, acceptance number, and rejection number are counts, and the risks are percentages. The reference below shows the Z1.4 sample size for general inspection level II across common lot sizes, the sample the standard prescribes before the AQL is applied. It is the fastest way to see how the sample scales with the lot.

Frequently asked questions

Sources, disclaimer, and editorial transparency

The sampling plans, code letters, and acceptance criteria used here follow ANSI/ASQ Z1.4 (and its equivalents ISO 2859-1 and MIL-STD-105), and the OC curve is computed from the binomial distribution. This calculator and guide are built and reviewed by the OpsCalculators team; see our Editorial Policy for how each tool is researched, built, and tested.

Results are accurate estimates for planning and education, not certified engineering advice, and this tool covers single sampling under normal inspection at general levels I to III; consult the full standard for tightened, reduced, double, or special-level plans. Validate against the published tables before making acceptance decisions. See our full Disclaimer. OpsCalculators.com is operated by MAFHH INTERNATIONAL LTD. Your inputs are processed in your browser and are never stored; see our Privacy Policy.