Cpk Calculator
Calculate Cpk and related process capability indices from your specification limits, process mean, and standard deviation.
Calculator is for informational purposes only. Terms and Conditions
Uses the standard normal-process capability relationship; a meaningful capability study also requires a stable, representative process.
Enter the process values
Use the same measurement units for specification limits, mean, and standard deviation.
For the default two-sided calculation, enter LSL, USL, process mean, and a positive process standard deviation.
Process capability result
Cpk is shown first, followed by the one-sided capability components and normal-model quality estimates.
Result details
- Cp—
Show calculation steps Review the capability ratios, limiting specification side, and normal-tail estimates
- Enter valid values to see the complete calculation.
Capability by specification side
Compare the applicable one-sided capability indices with the optional required-Cpk threshold. The smaller applicable capability value controls Cpk.
- Enter valid values to populate the chart.
Method, Sources, and Assumptions
Calculation basis, authoritative references, assumptions, and limitations for interpreting capability.
Cpk is calculated as the minimum applicable one-sided capability ratio. Cp is reported only for a two-sided specification. PPM and yield are theoretical normal-distribution estimates based on the entered mean and standard deviation.
- Specification limits are engineering/customer acceptance limits, not control-chart limits.
- Traditional Cp/Cpk interpretation assumes approximately normal data and a statistically stable process.
- The entered standard deviation materially determines the result; within-process and overall standard deviations can support different capability/performance interpretations.
- If the mean and standard deviation are estimated from a sample, capability-index uncertainty can be material; NIST notes that sufficiently large, independent samples are needed for reliable estimates.
- A selected required Cpk is a comparison criterion only and does not establish a universal acceptance standard.
Calculator guide
Understanding Your Cpk Result
Cpk measures the standardized distance from the process mean to the nearest applicable specification limit relative to process variation. The calculator above uses the applicable specification limit or limits, the process mean, and a positive process standard deviation to return Cpk; for two-sided specifications it also reports Cp, Cpu, and Cpl, plus normal-model PPM and yield estimates.
A larger Cpk means the process mean is farther from the nearest specification limit in standard-deviation units. That number is useful only when the data represent the process you intend to judge: NIST defines process capability in terms of an in-control process, and ASQ emphasizes stable, representative sampling before capability is projected to future production.
- Required inputs
- Specification limit(s), process mean, and process standard deviation.
- Primary output
- Cpk, a dimensionless index controlled by the weaker specification side.
- Key condition
- Use a stable, representative process and a standard-deviation estimate appropriate to the study.
Cpk Inputs and Outputs
The calculator accepts two-sided, upper-only, and lower-only specifications. All dimensional inputs must describe the same characteristic in the same units; the capability indices themselves are dimensionless.
- Lower Specification Limit (LSL)
- The lowest acceptable value defined by the product, drawing, customer, or engineering requirement. It is not the lower control limit from a control chart.
- Upper Specification Limit (USL)
- The highest acceptable value for the characteristic. A USL is an acceptance requirement, not a statistical control-chart boundary.
- Process mean
- The average value of the process data being evaluated. If the mean moves toward one specification limit, Cpk falls even when the process spread is unchanged.
- Process standard deviation
- The variation estimate used in the capability calculation. For classical Cpk interpretation, the choice of within-process versus overall variation is a critical methodological decision.
- Cpk
- The smaller applicable one-sided capability ratio. For a two-sided specification, the closer side of the process to its limit controls the result.
- Cp, Cpu, and Cpl
- Cp compares total specification width with process spread; Cpu measures capability to the upper limit; Cpl measures capability to the lower limit. Cp is meaningful only when both LSL and USL exist.
Cpk Formula and Calculation Method
Cpk evaluates the distance from the process mean to each applicable specification limit and divides that distance by three standard deviations. For a two-sided specification, the smaller side becomes Cpk.
Two-sided Cpk formula
Plain language: calculate the upper-side capability and lower-side capability, then use the smaller value.
NIST gives the same population form using \(\mu\) and \(\sigma\), and the sample-estimator form using \(\bar{x}\) and \(s\). The calculator accepts the mean and standard deviation you provide; deciding whether those estimates appropriately represent a stable process remains part of the capability study.
Cp, Cpu, and Cpl
Cp measures potential capability from total tolerance width and spread. Cpu and Cpl isolate the upper and lower sides; \(C_{pk}=\min(C_{pu},C_{pl})\).
- \(C_{pk}\)
- Process capability index Capability to the nearest applicable specification limit after accounting for centering.
- \(C_p\)
- Potential process capability Two-sided specification width divided by six standard deviations; it does not penalize an off-center mean.
- \(C_{pu}\)
- Upper capability index Distance from the process mean to the upper specification limit divided by three standard deviations.
- \(C_{pl}\)
- Lower capability index Distance from the lower specification limit to the process mean divided by three standard deviations.
- \(USL\)
- Upper specification limit Maximum acceptable value for the characteristic.
- \(LSL\)
- Lower specification limit Minimum acceptable value for the characteristic.
- \(\bar{x}\)
- Sample process mean Average of the measurements used to characterize the process.
- \(s\)
- Estimated process standard deviation Variation estimate entered into the sample capability equation.
Worked Cpk Example
The calculator’s default example uses a two-sided specification from 9.900 to 10.100, a process mean of 10.002, and a standard deviation of 0.022. The mean is slightly above the specification midpoint, so the upper side should control Cpk.
Calculate the one-sided capability values
Calculate Cp and select Cpk
Result
Cpk ≈ 1.485
The upper specification side controls because Cpu is smaller than Cpl. The result is above the example comparison threshold of 1.33, but that comparison does not establish whether 1.33 is the correct acceptance requirement for a real process.
How to Interpret Cpk, Cp, Cpu, and Cpl
Cpk answers two questions at once: is the process spread small enough relative to the specification, and is the process mean positioned far enough from the nearest limit? Comparing Cpk with Cp and the two one-sided indices separates those effects.
Cp much higher than Cpk
The tolerance width may be adequate in principle, but process centering is consuming capability. The smaller of Cpu and Cpl identifies the limiting side.
Cp close to Cpk
The process is relatively centered within the two-sided specification. If both indices are low, reducing spread matters more than simply moving the mean.
Negative Cpk
A negative one-sided ratio occurs when the process mean itself lies beyond the applicable specification limit. Verify the data and specification before treating the number as a routine capability result.
What counts as a good Cpk?
There is no single universal cutoff. NIST’s glossary describes a capable process using Cpk above 1.0 under one definition and notes that other promoted criteria include 1.33, 1.5, and 2.0. A real acceptance decision should use the requirement that actually governs the product or process.
| Cpk | Nearest limit distance | Interpretation |
|---|---|---|
| < 0 | Mean beyond a limit | The process mean is outside at least one applicable specification limit. |
| 1.00 | 3 standard deviations | The nearest limit is 3s from the mean; this is not automatically an adequate requirement for every application. |
| 1.33 | About 4 standard deviations | A commonly used comparison value in practice, but not a universal acceptance rule. |
| 1.67 | About 5 standard deviations | More margin to the nearest limit, subject to the same stability and distribution assumptions. |
| 2.00 | 6 standard deviations | Very large calculated separation from the nearest limit; verify data quality and the sigma estimate before assuming equivalent field performance. |
Do not turn Cpk directly into a universal defect rate
For a centered normal process, NIST gives reference reject rates for several Cp values, but explicitly notes that those reject figures assume the distribution is centered. When the mean is off center, the two tails are different. The calculator therefore estimates the lower and upper normal tails from the entered mean, standard deviation, LSL, and USL rather than using Cpk alone.
How to Improve Cpk
Improving Cpk means increasing the standardized distance from the process mean to the nearest specification limit. With the specification limits held constant, there are two mathematical levers: move the mean away from the limiting specification side, reduce process variation, or do both.
When Cp is much higher than Cpk
The process has more potential capability than its current centering allows. The smaller of Cpu and Cpl identifies the limiting side. Investigate sources of process bias such as setup offset, machine adjustment, fixture alignment, calibration, tool compensation, or a target setting before assuming the specification itself should change.
When Cp and Cpk are both low
The process spread is consuming most of the available tolerance. Investigate variation from equipment condition, tooling, material, operating parameters, environment, operator method, and the measurement system. A change that lowers standard deviation improves both Cp and Cpk when the mean and limits stay fixed.
Controlled sensitivity check using the calculator example
In the default example, Cpk is about 1.485 with a mean of 10.002 and a standard deviation of 0.022. If the process mean is moved to the 10.000 specification midpoint while the limits and standard deviation stay unchanged, the two sides become equal and Cpk rises to Cp:
That is a modest improvement because the example process is already close to centered. If instead the mean stays at 10.002 and standard deviation is reduced by 10%, from 0.022 to 0.0198, the upper side remains limiting but:
With the mean and specification limits held constant, Cpk is inversely proportional to the standard deviation. A 10% reduction in standard deviation therefore increases Cpk by about 11.1% in this controlled example. This relationship describes the mathematics; the process change needed to achieve that reduction must be established from process data.
Cpk vs. Ppk: Which Standard Deviation Matters?
The formula shape for Cpk and Ppk is similar; the important difference is the variation estimate placed in the denominator. Calling every standard deviation “sigma” without defining how it was estimated can change the meaning of the index.
Cpk: within-process capability
ASQ describes capability estimation using pooled within-group standard deviations from observations taken close together in time. That approach is intended to characterize inherent within-process variation without folding process-mean shifts between sampling periods into the spread estimate.
Ppk: overall observed performance
ASQ’s Certified Quality Process Analyst handbook defines Pp using the sample standard deviation when random and special causes may both be present and states that statistical control is not required; Ppk is the smaller upper or lower process-performance index. In practice, this overall-spread basis is what distinguishes the performance interpretation from classical within-process Cpk.
| Question | Cpk | Ppk |
|---|---|---|
| Variation basis | Within-process or within-subgroup estimate | Overall observed variation across the study period |
| What it emphasizes | Capability of a stable process under its inherent variation | Actual combined performance over the observed period |
| Effect of long-term shifts | Designed to reduce their effect when within variation is estimated properly | Included in the overall spread |
| If Cpk is much higher than Ppk | Investigate drift, shifts, subgroup-to-subgroup changes, lot effects, setup changes, or other time-dependent variation rather than assuming the difference has one cause. | |
Data Quality, Stability, Normality, and Sample Size
A precise-looking Cpk is only as trustworthy as the process data behind the mean and standard deviation. Stability, representative sampling, distribution shape, sample size, and measurement quality can all change how much confidence you should place in the result.
- Check process stability first. NIST frames capability as a comparison between an in-control process and specification limits, and ASQ states that sampling from a stable system is essential for meaningful estimates of future process performance.
- Use representative data. A short run from one machine state, one cavity, one shift, or one material lot may miss sources of variation that matter to the decision.
- Check the distribution assumption. NIST states that conventional Cp and Cpk assume normally distributed process values; ASQ notes that capability estimates are highly sensitive to that assumption and suggests transformations or alternative distributions for stable non-normal processes.
- Treat sample estimates as uncertain. NIST states that capability-index estimates generally need a sufficiently large sample and describes about 50 independent values as a general “large enough” rule of thumb, while also noting that capability studies often need 100 or more observations and that confidence-interval variation is not negligible.
- Verify the measurement system. Measurement resolution, calibration, repeatability, reproducibility, fixture effects, and inspection method can add apparent variation or hide real variation. Cpk does not separate process variation from measurement error automatically.
- Keep units consistent. LSL, USL, mean, and standard deviation must use the same physical unit. Cpk itself is dimensionless, so converting every dimensional input consistently should leave the index unchanged.
Common Cpk Mistakes
Most serious Cpk errors come from misidentifying the data or the limits, not from the arithmetic itself.
Using control limits as specification limits
USL and LSL come from acceptance requirements. UCL and LCL are statistical limits derived from process behavior. Replacing one with the other changes the question being answered.
Ignoring an unstable process
A single mean and standard deviation can hide drift, cycles, step changes, and special causes. Stabilize or appropriately model the process before treating Cpk as predictive.
Using the wrong standard deviation
Overall variation and within-process variation can produce different indices. Record the sigma-estimation method and call the resulting metric Cpk or Ppk consistently.
Assuming Cpk = 1.33 is universal
The calculator’s 1.33 value is an editable example threshold. Acceptance criteria must come from the applicable requirement rather than from a generic web table.
Converting Cpk directly to defects
Defect probability depends on the actual distance from the mean to each limit and on the distribution model. Off-center processes have unequal tails even when one Cpk value summarizes the limiting side.
Rounding at a decision boundary
If an acceptance threshold is 1.33, do not decide from a heavily rounded display when the unrounded estimate is close to the boundary. The calculator can display additional significant figures, but more digits do not remove sampling uncertainty.
Assumptions and Limits of Cpk
Cpk is a compact statistical index, not a complete process-validation method. The formula cannot determine whether your data collection, measurement system, distribution model, or acceptance requirement is appropriate.
Stable-process assumption
Classical capability interpretation assumes the process is sufficiently stable for the estimated mean and variation to characterize the state you intend to project.
Normal-model assumption
Traditional Cp/Cpk and the calculator’s theoretical PPM/yield estimates use a normal-distribution framework. Material skew, heavy tails, mixtures, truncation, or multimodality can make normal-tail estimates misleading.
Point-estimate uncertainty
The sample mean and standard deviation vary from sample to sample, so Cpk also varies. A point estimate near a requirement may need a confidence interval or lower confidence bound.
Specification validity
The calculator assumes the LSL and USL you enter are the correct engineering acceptance limits for the characteristic. It cannot determine whether the drawing, customer requirement, or specification is current.
One-sided capability
When only an upper or lower limit exists, the calculator uses Cpu or Cpl as the applicable one-sided capability result. Cp is omitted because total specification width is undefined without both limits.
Measurement-system effects
Observed spread contains any measurement variation present in the data. The calculator does not perform a measurement-system analysis or remove gauge variation.
Sources and Calculation Checks
The formulas and capability-study limitations in this guide were checked against NIST and ASQ. The worked example was recomputed independently from both the one-sided ratios and the direct Cp relationship, then cross-checked against the calculator’s default state.
- NIST/SEMATECH Engineering Statistics Handbook — What is Process Capability? — supports Cp, Cpk, Cpu, Cpl formulas; one-sided capability; normality and stability context; sample-size discussion; confidence intervals; and centered-process reject-rate examples.
- NIST/SEMATECH Engineering Statistics Handbook — Glossary — supports the Cpk definition for stable processes and notes that different Cpk criteria, including 1.0, 1.33, 1.5, and 2.0, are used in practice.
- ASQ — Process Capability — supports representative sampling, pooled within-group variation, sensitivity to normality, the need for a stable system, confidence-interval context, and alternative approaches for stable non-normal data.
- ASQ Certified Quality Process Analyst Handbook — Pp/Ppk glossary definitions — supports the process-performance definitions of Pp and Ppk and the use of overall sample variation when statistical control is not assumed.
- ASQ Quality Progress — What Is 3.4 per Million?— supports the distinction between a centered six-standard-deviation process and the 3.4-per-million convention associated with an assumed 1.5-sigma mean shift.
How the example was checked: Cpu and Cpl were calculated independently, their average was compared with direct Cp, and the limiting side was confirmed from the mean’s position relative to the specification midpoint. The normal-tail PPM check used the entered mean, standard deviation, and each specification limit separately rather than converting Cpk alone into a defect rate.
Cpk Calculator FAQ
These answers address the most important follow-up questions when a Cpk number needs to be interpreted rather than merely calculated.
What is a good Cpk value?
There is no universal value that is “good” for every process. NIST notes multiple criteria used in practice, including 1.0, 1.33, 1.5, and 2.0. Use the requirement that applies to the customer, product, process, or quality plan, and verify the process is stable before making the comparison.
Can Cpk be negative?
Yes. If the process mean lies above the USL, Cpu is negative; if the mean lies below the LSL, Cpl is negative. Because Cpk uses the smaller applicable side, the resulting Cpk can be negative.
Can Cpk be greater than Cp?
Not under the conventional two-sided definitions using the same mean and standard deviation. NIST shows that Cpk is less than or equal to Cp; equality occurs when the process is centered at the specification midpoint.
What is the difference between Cpk and Ppk?
The key difference is the variation estimate. Cpk is generally associated with within-process or within-subgroup variation for a stable process, while Ppk uses overall observed variation across the study period. If you enter an overall standard deviation into this calculator, the arithmetic is the same form but the interpretation is closer to Ppk.
How many samples do I need for Cpk?
There is no universal sample count that guarantees a valid study. NIST describes about 50 independent observations as a general “large enough” rule of thumb for capability-index estimates, while also noting that capability studies often need 100 or more observations and that confidence-interval variation matters. Sampling design and process stability are as important as the count.
Does Cpk require normally distributed data?
Traditional Cp/Cpk interpretation assumes a normal population. For a stable but materially non-normal process, ASQ notes that transformation or an alternative probability distribution may be more appropriate. The calculator’s PPM and yield estimates specifically use a normal model.
Can I calculate Cpk with only one specification limit?
Yes. NIST defines Cpu for an upper-only specification and Cpl for a lower-only specification. The calculator’s Advanced Options let you select either one-sided case. Cp is not reported because total specification width requires both an LSL and a USL.
Why can a process have high Cp but lower Cpk?
Cp measures the available two-sided tolerance relative to total spread but does not penalize an off-center mean. Cpk uses the closer specification side, so a shifted mean can lower Cpk even when the process spread is small enough to produce a high Cp.
What does a Cpk of 1.33 mean?
A Cpk of 1.33 places the nearest applicable specification limit about \(3\times1.33=3.99\), or approximately four standard deviations, from the process mean under the standard-deviation definition used in the calculation. It does not prove that the process is stable, and 1.33 is not a universal acceptance requirement.
Does Cpk = 2.0 mean Six Sigma?
For a centered two-sided normal process, \(C_p=C_{pk}=2.0\) places each specification limit six standard deviations from the mean. That centered relationship should not be confused with the widely cited Six Sigma value of 3.4 defects per million opportunities, which ASQ describes in connection with an assumed 1.5-sigma long-term mean shift.