Process Capability Calculator

What process capability tells you, and why it matters

Process capability answers a question every manufacturer eventually has to face: can this process reliably make product inside the limits the customer set? It is a comparison of two things. The first is the natural spread of the process, how much the output varies when nothing unusual is happening, conventionally taken as six standard deviations wide and called the voice of the process. The second is the tolerance, the band between the lower and upper specification limits, called the voice of the customer. Capability indices put a single number on how comfortably the first fits inside the second, so a team can tell at a glance whether a process is safe, marginal, or bound to produce scrap.

The two headline indices are Cp and Cpk. Cp measures only whether the spread fits: it is the tolerance width divided by six standard deviations, so a Cp of 1.0 means the spread exactly fills the tolerance and a Cp of 2.0 means the tolerance is twice as wide as the spread.

But Cp ignores where the process is centered, and a process can have plenty of room yet still make defects if it drifts to one side. Cpk fixes that by measuring the distance from the process mean to the nearer specification limit in units of three standard deviations, then reporting the worse of the two sides.

When a process is centered, Cp and Cpk agree; when it is off-center, Cpk falls below Cp, and the gap tells you exactly how much re-centering would buy.

This calculator computes all four common indices, Cp, Cpk, Pp, and Ppk, along with the process sigma level, defects per million, and expected yield. You can paste raw measurements and let it work out the statistics and draw a histogram, or enter a mean and standard deviation you already have. It uses the true normal distribution for the defect and sigma figures rather than a rounded table, and it handles one-sided specifications, so a single tool covers the full range of everyday capability questions on the shop floor.

How this calculator works, step by step

Start by choosing how you will provide the numbers. In the default data mode you paste your measurements, one per line or separated by commas, and the tool computes the mean, the overall standard deviation across all the readings, and a short-term standard deviation estimated from the variation between consecutive points. In summary mode you instead type a mean and a standard deviation you already have, and you can optionally add a separate short-term standard deviation if you want distinct short- and long-term indices; with a single standard deviation the short- and long-term indices are equal.

Next enter the specification limits. Provide both a lower and an upper limit for a two-sided tolerance, which yields Cp and Cpk, or leave one blank for a one-sided characteristic, which yields a one-sided Cpk. The tool validates that the upper limit is above the lower one and that you have supplied at least one. A checkbox controls whether the reported process sigma includes the conventional 1.5-sigma long-term shift, so you can quote either the short-term sigma from a capability study or the shifted figure that lines up with the familiar 3.4 defects per million at six sigma.

The result panel then shows the full picture. The headline is Cpk, the index most teams act on, and below it sit Cp, Cpk, Pp, and Ppk together with the mean and both standard deviations, so the short-term and long-term views are visible side by side. The process sigma, DPMO, expected yield, and parts per million out of specification translate the same result into the language of Six Sigma. In data mode the chart draws a histogram of your readings so you can see the shape of the distribution; in summary mode it draws the fitted normal curve. Download a PDF or CSV or share the result; everything runs in your browser and nothing you enter is stored.

The formulas behind the indices

The arithmetic rests on comparing spread to tolerance. Cp is the tolerance width divided by six standard deviations, (USL minus LSL) over six sigma, capturing potential capability with no regard to centering. Cpk takes the distance from the mean to each specification limit, divides each by three standard deviations, and reports the smaller: the minimum of (USL minus the mean) and (the mean minus LSL), all over three sigma. Because it uses the nearer limit, Cpk can never exceed Cp, and equals it only when the process is perfectly centered.

The difference between the C indices and the P indices is entirely the standard deviation used. Cp and Cpk use the within-subgroup, short-term standard deviation, which reflects only the variation present within a short burst of production; Pp and Ppk use the overall, long-term standard deviation of all the data, which also includes the drift and shifts that accumulate over time.

From pasted data this tool estimates the short-term standard deviation with the average moving range divided by 1.128, the standard individuals-chart constant, and the long-term standard deviation with the ordinary sample standard deviation.

The defect figures come from the normal distribution: the probability of falling beyond each specification limit is the normal tail area at that limit, and DPMO is that total probability times a million, with expected yield its complement. The process sigma is the normal score corresponding to the yield, optionally plus 1.5.

Five worked examples you can follow

Example 1: a centered process at Cpk 1.0

Suppose a shaft diameter has a mean of 10.00 mm, a standard deviation of 1.00 mm, and specification limits of 7 and 13. The tolerance width is 6, so Cp is 6 divided by six times one, exactly 1.0. The mean sits dead center, 3 mm from each limit, so Cpk is 3 divided by three times one, also 1.0. A centered process at Cp equal to Cpk equal to 1.0 corresponds to about a 3-sigma process, roughly 2,700 defects per million and a 99.73 percent yield. This is the classic barely-capable process: the spread just fills the tolerance, leaving no margin for drift.

Example 2: the same spread, off center

Now shift the mean to 11.00 mm with the same standard deviation and limits. Cp is unchanged at 1.0 because the spread has not changed. But the mean is now only 2 mm from the upper limit and 4 mm from the lower, so Cpk is the smaller side, 2 divided by three times one, about 0.67. Defects climb to roughly 23,000 per million, almost all beyond the upper limit. The lesson is stark: without changing variation at all, letting the process drift off center has nearly ten-timesed the defect rate, and the fix is re-centering, not tightening.

Example 3: reaching the 1.33 target

Take a fill weight centered at 500 g with specification limits of 494 and 506, a tolerance width of 12. To reach the common Cpk target of 1.33, the standard deviation must satisfy 6 divided by six sigma equal to 1.33, so sigma must be about 0.75 g. At that spread the process runs at roughly a 4-sigma level, about 63 defects per million. If the measured standard deviation is larger, say 1.0 g, Cp falls to 1.0 and the process misses the target; the gap tells the team how much variation reduction, here about 25 percent, is needed to qualify the process.

Example 4: a one-sided specification

Some characteristics have only one limit. Consider surface roughness with an upper limit of 3.2 micrometers, a mean of 2.0, and a standard deviation of 0.4. There is no lower limit, so Cp is not defined; the one-sided Cpk is (3.2 minus 2.0) divided by three times 0.4, which is 1.2 divided by 1.2, exactly 1.0. The interpretation is the same as a two-sided Cpk of 1.0 on that side: about a 3-sigma margin to the single limit, with the defect rate driven entirely by the upper tail.

Example 5: short-term versus long-term with real data

Paste twenty readings that average 10.0 with a within-point standard deviation, from the moving range, of about 0.7 but an overall standard deviation of about 0.85 because the mean drifted during the run. With limits of 7 and 13, Cpk from the short-term sigma is about 1.43 while Ppk from the overall sigma is about 1.18. The process has good short-term potential but loses roughly a fifth of it to drift over time. That gap is the signal to hunt for the source of the drift, tool wear, temperature, or material lots, because closing it would lift long-term performance to match the short-term promise.

Three expert tips for a trustworthy capability study

Short-term and long-term: the story Cpk and Ppk tell together

The most useful thing this calculator does is show the short-term and long-term indices side by side, because the relationship between them is diagnostic. Cpk, built on the within-subgroup standard deviation, describes the process at its best: the variation present over a short window when only common causes are acting.

Ppk, built on the overall standard deviation of all the data, describes the process as the customer actually experiences it over weeks and months, including every shift change, tool adjustment, and material lot. When the two are close, the process is consistent and its short-term promise holds up over time.

When Cpk is much higher than Ppk, the process drifts, and its long-term output is worse than a snapshot would suggest.

That gap is a specific, actionable signal rather than a vague warning. It says the variation reduction that would most improve the customer’s experience is not about tightening the moment-to-moment scatter, which is already good, but about controlling the between-subgroup drift: finding why the mean wanders and holding it steady.

Common culprits are tool wear that shifts the center over a run, temperature cycles across a day, and differences between raw-material batches. A process with a high Cpk and a low Ppk is often easier to improve than one with a uniformly low Cpk, because the potential is already there and the task is to stop losing it.

Reporting only one index hides this; showing both turns a single capability study into a piece of improvement direction.

Reading the process sigma, DPMO, and yield

Capability indices and the Six Sigma metrics are two languages for the same thing, and this calculator prints both so a result can travel between an engineering review and a management dashboard without translation. The process sigma level expresses how many standard deviations of margin the process holds to its limits; DPMO counts the defects per million opportunities the current spread and centering imply; and expected yield is simply the complement, the fraction of output that lands in spec. All three come from the same normal-tail calculation behind the indices, so they always agree.

The one subtlety worth understanding is the 1.5-sigma shift. A short-term capability study captures the process at a moment, but real processes drift over the long haul, and empirical work found that long-term performance is typically about 1.5 sigma worse than the short-term snapshot suggests.

The Six Sigma convention bakes that in, which is why a process that looks like six sigma in a short study is quoted at 3.4 defects per million, the number you would get at 4.5 sigma. This tool lets you toggle the shift, so you can report the honest short-term sigma from your data or the shifted figure that matches published Six Sigma tables and the expectations of an audience trained on them.

Knowing which one a number refers to prevents the common confusion of comparing a short-term sigma from one report against a shifted sigma from another.

Capability, stability, and the order they come in

A capability index is only trustworthy for a process that is already stable, and getting that order right is the single most common way capability studies go wrong. Stability, or statistical control, means the process varies only from common causes, the ordinary background noise, with no special causes, sudden shifts, trends, or outliers, disturbing it. A stable process is predictable: its distribution tomorrow will look like its distribution today, so an index computed now describes future output too. An unstable process has no fixed distribution, so its capability index is a snapshot of something that will not hold, a number with a false air of precision.

That is why the discipline pairs control charts with capability and always in the same sequence: chart first, capability second. The control chart is the tool that decides whether special causes are present, by testing the data against limits computed from the process itself. Only once it shows the process is in control does the capability index become a valid prediction.

Computing Cpk on an out-of-control process is a classic error; the resulting number may look fine or terrible, but either way it does not describe a stable reality.

When this calculator’s histogram shows obvious outliers, multiple peaks, or a clear trend across pasted data in run order, treat the capability figures with caution and investigate stability before acting on them.

When the normal assumption does not hold

The standard capability indices and the DPMO conversion assume the measurements follow a normal, bell-shaped distribution, and for a great many characteristics that assumption is close enough to be useful. Dimensions like length, diameter, and weight, driven by many small independent sources of variation, tend toward normal by the central limit theorem, and for them the indices this tool reports are accurate enough to drive decisions. The convenience of the normal model, closed-form indices and a clean map to sigma and DPMO, is well earned when the data cooperates.

The assumption strains for characteristics that are naturally skewed or bounded. Flatness, roundness, concentricity, and other geometric characteristics cannot go below zero and pile up against that boundary, producing a one-sided, skewed distribution that a symmetric bell curve describes poorly. Applying a normal-based Cpk there can badly misstate the defect rate, usually understating it.

The remedies are to transform the data to approximate normality, for example with a Box-Cox transformation, or to use a non-normal capability method that fits the actual distribution and computes the indices from its true tails.

The practical guidance is to always look at the histogram this calculator draws: if it is roughly symmetric and single-peaked, trust the indices; if it is strongly skewed, piled against a limit, or clearly multi-modal, treat the figures as approximate and reach for a distribution-appropriate method before making a capability claim.

Common mistakes in capability analysis

A handful of errors account for most misleading capability numbers. Watch for these before you act on an index.

• Skipping the stability check. Computing capability on an out-of-control process gives a precise-looking number that describes nothing stable. Chart first.
• Confusing Cpk and Ppk. They use different standard deviations and answer different questions. Quoting one as the other overstates or understates real performance.
• Ignoring the 1.5 shift. Comparing a short-term sigma against a shifted long-term sigma from another source mixes two scales and misleads by 1.5 sigma.
• Assuming normality blindly. Skewed or bounded data breaks the indices. Check the histogram before trusting Cpk on geometric or one-sided characteristics.
• Too little data. A handful of points gives an unstable estimate of the standard deviation and therefore of every index. Use enough readings, ideally 25 or more, spread over the real range of conditions.
• Chasing Cp when Cpk is the problem. If the gap between them is large, the process is off-center; re-center it rather than launching a variation-reduction project.
• Measurement error left unchecked. If the gauge consumes much of the tolerance, the observed spread is partly measurement noise, and the capability index is optimistic or pessimistic depending on the error. Validate the measurement system first.

Where process capability fits in the quality toolkit

Capability sits at the center of statistical process control, drawing on the tools before it and feeding the decisions after it. Before a capability study can be trusted, a control chart must confirm the process is stable, and the measurement system behind the readings must be validated so the spread you see is real rather than gauge noise. Capability then translates the stable, well-measured process into a single verdict, can it meet spec, that the Six Sigma metrics, sigma level, DPMO, and yield, restate for a wider audience. This page computes those metrics alongside the indices so the whole picture appears in one view.

From there capability drives action. A process below target routes to improvement: variation reduction if Cp itself is low, or re-centering if the gap between Cp and Cpk is large. A capable process supports decisions about how much to inspect, since a high, stable capability can justify reduced sampling, and about which characteristics need continued charting.

The other calculators in this quality silo, control chart limits to confirm stability, sigma level and DPMO to report performance, process yield to roll capability up across steps, acceptance sampling to set inspection, and Gage R&R to validate the gauge, all connect to the capability study at the heart of the workflow.

Return to the Quality Control hub for the full set as each launches.

A note on data quantity and sampling

The reliability of every index on this page depends on how much data went into it and how that data was collected, because capability is ultimately an estimate of a distribution, and estimates from small or biased samples are shaky. A capability figure from ten points carries wide uncertainty: the standard deviation, which sits in the denominator of every index, is poorly estimated from few readings, so the Cpk you compute could easily be off by a large margin in either direction. Most standards recommend at least 25 to 30 measurements for an initial study, and more when the stakes are high, precisely so the standard-deviation estimate settles down and the index means what it says.

How the data is gathered matters as much as how much. To capture the long-term variation that Ppk is meant to reflect, the sample must span the real range of operating conditions, different shifts, operators, material lots, and times of day, not a single convenient half-hour of production.

A study run entirely within one stable burst will show an impressive Ppk that collapses the moment normal variation reasserts itself, because it never saw that variation. The most honest capability studies collect subgroups spread across a representative window, which lets the within-subgroup and overall standard deviations differ meaningfully and gives the Cpk-versus-Ppk comparison its diagnostic power.

When you paste data here, paste a representative sample, not a cherry-picked good run, so the indices describe the process you actually have.

Turning a capability number into a decision

A capability index is only useful once it changes what someone does, and the decision it drives depends on where the number lands and why. A Cpk comfortably above the target, say 1.67 or more, says the process has margin to spare: the practical questions become whether inspection can be reduced, whether the tolerance is tighter than it needs to be, and whether attention is better spent on a weaker process elsewhere. A Cpk near the target of 1.33 says the process is acceptable but has little cushion, so it warrants continued charting and a watchful eye on the sources of drift, because a small shift could push it below the line.

A Cpk below target is the case that demands action, and the first diagnostic is the gap between Cp and Cpk. When Cp is healthy but Cpk is low, the spread is fine and the process is simply off-center, so the fix is an adjustment to the mean, often quick and cheap, that can lift Cpk close to Cp.

When Cp itself is low, the spread is genuinely too wide for the tolerance, and no amount of re-centering will help; the response is a variation-reduction effort aimed at the dominant sources of scatter, or, where justified, a conversation about whether the tolerance is realistic.

Reading the two indices together turns a single failing number into a specific choice between two very different improvement paths, which is why this calculator always shows them side by side rather than reporting Cpk alone.

A short history of the indices

Capability indices grew out of the spread of statistical quality control from its origins at Bell Labs, where Walter Shewhart formalized the control chart in the 1920s and separated common-cause from special-cause variation. Once processes could be held stable, the natural next question was whether a stable process was good enough for its tolerances, and the ratio of tolerance to process spread, Cp, emerged as the answer. Cpk, adding the centering term, followed as practitioners noticed that a wide-enough spread was no guarantee of conforming product if the process ran off to one side.

The indices gained their modern prominence through two channels. The automotive industry, through what became the AIAG statistical process control manual, made Cp, Cpk, Pp, and Ppk a standard requirement in supplier quality, which is why they appear on production part approval paperwork across manufacturing.

In parallel, Motorola’s Six Sigma program in the 1980s recast the same idea on the sigma scale, tying capability to defects per million and popularizing the 1.5-sigma shift.

The two vocabularies, capability indices and sigma levels, describe the same underlying comparison of spread to tolerance, which is why a tool that reports both, as this one does, speaks to audiences trained in either tradition.

Variable data versus attribute data

The indices on this page apply to variable data, measurements on a continuous scale such as length, weight, temperature, or time, where each part yields a number and the spread of those numbers can be compared against limits. This is where Cp and Cpk live, because they depend on a standard deviation, which only exists for measured quantities. Whenever you can record an actual value rather than a pass or fail, variable-data capability is the richer choice: it detects drift toward a limit before any part fails, and it needs far fewer samples than counting defectives to reach a confident conclusion.

Attribute data, by contrast, records only a category, conforming or not, pass or fail, number of blemishes, and cannot produce a standard deviation, so classical Cp and Cpk do not apply. Capability for attribute data is expressed instead through the defect rate directly, as a proportion defective or defects per unit, which maps to a sigma level and DPMO through the same normal conversion this tool uses.

That is the bridge between the two worlds: a process measured with variable data gives Cpk, which converts to DPMO; a process tracked with attribute data gives DPMO directly, which converts back to an equivalent sigma level.

When you have the choice, measuring variables and computing capability is more informative than counting attributes, because it uses the distance to the limit, not just whether the limit was crossed.

Cpm and target-based capability

Cp and Cpk treat the whole tolerance band as equally good, but for some characteristics being on target matters more than merely being in spec, and a variant index, Cpm, captures that. Cpm is built on the Taguchi idea that quality loss grows as a characteristic moves away from its target, not just when it crosses a limit, so it penalizes any deviation of the mean from the target value. Where Cpk asks how close the process is to the nearer limit, Cpm asks how close it is to the target, folding both the spread and the distance of the mean from target into a single number. A process centered exactly on target has Cpm equal to Cp; as the mean drifts from target, Cpm falls faster than Cpk.

The distinction matters most for characteristics with a strong best value, a mating dimension that fits ideally at one size, a fill weight where both overfilling and underfilling cost money, an electrical parameter tuned to a nominal. For those, a process that meets spec but sits away from target is still losing quality in the Taguchi sense, and Cpk alone would call it fine.

This calculator reports the classic Cp, Cpk, Pp, and Ppk that the great majority of specifications and supplier requirements are written around; where a target-based view is needed, the mean and standard deviation it reports are exactly the inputs a Cpm calculation needs, and the histogram shows at a glance whether the process is centered on its target or merely inside its limits.

Units and quick reference

Enter measurements and specification limits in the same units; the indices are dimensionless ratios, the process sigma is unitless, and DPMO and yield are counts and percentages. The reference below shows how capability, the sigma level, and the defect rate line up for a centered process, using the short-term sigma without the 1.5 shift. It is the fastest way to translate a Cpk into what it means for defects and yield.

Frequently asked questions

Sources, disclaimer, and editorial transparency

The capability index formulas, the within- and overall-standard-deviation methods, and the sigma-to-DPMO conversion used here follow recognized quality-engineering sources, including the ASQ body of knowledge, the AIAG SPC reference manual, and standard statistical process control texts. 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. Confirm process stability and validate your measurement system before acting on any capability index. 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.