A Cpk of 1.33 places the process mean approximately 3.99 within-process standard deviations from the nearer specification limit: 3 × 1.33 = 3.99. This is often described as approximately four-sigma capability. It is a capability estimate under the study’s assumptions, not proof that the process is stable or that every unit meets specification.
A Cpk of 1.33 indicates a stable process that meets customer specifications but still has room for improvement, as variability remains present. Aiming for a higher Cpk, such as 2.0, would reduce defects and enhance product quality through continuous improvement efforts.
| Cpk | Sigma Level | Approx. DPMO | Interpretation |
|---|---|---|---|
| 1.33 | 4σ | ~63 DPMO | Common capability benchmark |
When you encounter a Cpk of 1.33, it signals that your process is generally capable but still has room for improvement. Interpret this estimate against the customer’s requirements only after independently checking stability, measurement quality and the study assumptions. Understanding what this means for your operations is crucial. What steps can you take to enhance your Cpk process capability and elevate quality? Let’s explore the implications and strategies together.
What Cpk 1.33 Does—and Does Not—Establish
Check the measurement system, process stability, sampling, and distribution model before relying on the result. Customer requirements may call for a different minimum Cpk. A centered normal model with limits approximately four standard deviations from the mean gives roughly 66 out-of-limit units per million using 3.99, or roughly 63 using an exact four-sigma approximation. Those are modeled, unshifted estimates; they are not a universal defect prediction for any process with Cpk 1.33.
Do not add a 1.5-sigma shift to that interpretation without explicitly explaining the different convention. Off-centering, nonnormality, and changes over time can alter defect rates.
How to Interpret the 1.33 Benchmark
A Cpk of 1.33 is an estimate of capability relative to the nearer specification limit. The useful question is whether that result meets the requirement for this characteristic under the conditions studied. It does not establish statistical control, guarantee that every unit meets specification, or replace the customer’s acceptance criteria.
For the general formula, study design and manufacturing examples, use our Cpk process capability guide. This page focuses on interpreting the 1.33 result.
Why 1.33 Is Often Described as Four-Sigma Capability
The distance from the mean to the nearer limit is approximately 3 × Cpk within-process standard deviations. At 1.33, that distance is 3.99 standard deviations, commonly rounded to four. A defect-rate interpretation needs additional distribution and centering assumptions. Do not mix this distance with a convention that adds a long-term mean shift.
A Worked Cpk 1.33 Example
Suppose a stable process has a lower specification of 9.6, an upper specification of 10.4, a mean of 10.0 and an estimated within-process standard deviation of 0.1, all in the same measurement unit. Each specification is 0.4 from the mean.
Cpk = 0.4 ÷ (3 × 0.1) = 1.333…, or 1.33 when rounded. If the mean shifts while the spread remains the same, the distance to one limit decreases and Cpk falls. This illustrates why a previously acceptable result needs continued monitoring.
Comparing 1.33 With Other Targets
A result of 1.0 corresponds to approximately three standard deviations to the nearer limit; 1.33 corresponds to approximately four; 1.67 to approximately five; and 2.0 to six. These comparisons describe the model’s distance from the limit. They are not universal acceptance categories or industry guarantees.
Keep the acceptance target separate from the measured estimate. A characteristic with serious failure consequences may need a different target from a routine one. Document the applicable requirement and consider the uncertainty of the estimate before accepting a marginal result.
What to Check Before Accepting a 1.33 Result
Confirm the specification limits and measurement units, check the measurement system, and establish that the sampled observations represent the process conditions. Review stability and the distribution model before interpreting capability. Stratify by machine, cavity, material or shift when combining unlike streams would hide important differences.
Use enough representative data for the study’s purpose. There is no single sample count that makes every capability study reliable. A result rounded to 1.33 can sit close to an acceptance boundary, so follow the agreed rule for rounding and confidence limits.
Improving a Result Below the Required Benchmark
If the process has adequate potential spread but is off center, investigate the cause of the shift. If its spread is too large, investigate the sources of variation. Stabilize special causes before treating a capability estimate as representative of ongoing operation.
Compare Cp and Cpk to understand centering, and review Ppk and overall process performance when the study spans changing conditions. Recalculate after improvement using a defined, representative dataset.
Frequently Asked Questions About Cpk 1.33
Does Cpk 1.33 prove that a process is stable?
No. Stability is assessed separately, typically with an appropriate control chart. A capability estimate does not establish that the process remained stable over time.
What sigma level corresponds to Cpk 1.33?
Multiplying 1.33 by 3 gives 3.99. Under the capability model, the process mean is approximately four within-process standard deviations from the nearer specification limit. Any defect-rate estimate depends on distribution, centering and other assumptions.
Is Cpk 1.33 always an acceptable target?
No. A customer, specification or risk assessment may require a different target. Confirm the applicable acceptance rule, sampling plan and confidence requirements.
What should a team do if Cpk is below 1.33?
First verify the data, measurement system and process stability. Then investigate centering and variation, make a controlled improvement and recalculate using representative data.