A process's sigma level is a single number that says how many standard deviations fit between its average and the nearest specification limit, a rough measure of how much room the process has before it makes a defect. Higher is better: a 3-sigma process yields about 66,800 defects per million, while a 6-sigma process yields 3.4, thanks to a 1.5-sigma shift built into the standard tables.
The number is useful because it turns "our quality is pretty good" into something you can compare across lines, plants, and products. But it is also widely misread, mostly because of that 1.5-sigma shift that lets "six sigma" mean 3.4 defects per million rather than the two-per-billion the pure statistics would give. This post explains what a sigma level actually measures, the yield each level delivers, where the 1.5-sigma shift comes from, and how to estimate your own.
What does a sigma level actually measure?
A sigma level measures how many standard deviations you can fit between the process average and the closest specification limit. Standard deviation, written with the Greek letter sigma, is the natural yardstick of a process's spread. If the distance from your average to the nearest limit is six standard deviations, you have a lot of buffer and very few parts fall outside the spec; if it is only three, the tail of the distribution pokes well past the limit and defects climb fast.
Two things follow from that picture. First, sigma level depends on both spread and centering: a tight process centered in the middle of the spec scores high; the same tight process drifting toward one limit scores lower, because one tail crept closer. Second, sigma level is close kin to process capability, the Cpk index measures the same buffer on a different scale, and a Cpk of 2.0 corresponds to a centered six-sigma process. They are two dialects for the same idea.
What yield does each sigma level give?
Each sigma level maps to a defect rate and a yield, and the jumps between levels are enormous, moving up one sigma cuts defects by roughly a factor of ten or more. The standard table below already includes the 1.5-sigma shift explained in the next section, which is why the numbers are the ones quoted in most Six Sigma material.
| Sigma level | Defects per million (DPMO) | Yield |
|---|---|---|
| 2 sigma | ~308,500 | ~69.1% |
| 3 sigma | ~66,800 | ~93.32% |
| 4 sigma | ~6,210 | ~99.379% |
| 5 sigma | ~233 | ~99.977% |
| 6 sigma | 3.4 | ~99.99966% |
The table makes two realities concrete. A three-sigma process, which many operations quietly run at, throws about 66,800 defects per million, roughly a 6.7% defect rate. That can be invisible day to day and ruinous at scale. And the climb is punishing: getting from five to six sigma means cutting an already-tiny 233 defects per million down to 3.4. The cost of quality is what decides whether that last step is worth it, and for most product characteristics it is not, you spend six-sigma money only where a defect is catastrophic or the volume is astronomical.
Where does the 1.5-sigma shift come from?
The 1.5-sigma shift is an adjustment, originated at Motorola in the 1980s, that assumes a process center will drift by about 1.5 standard deviations over the long run even when it is well controlled in the short run. Motorola engineers, the effort is associated with Bill Smith around 1986, observed that processes measured as tightly centered in a short study tend to wander over weeks and months as tools wear, temperatures swing, materials vary, and operators change. Rather than pretend that drift away, they baked a 1.5-sigma allowance into the conversion tables.
This is why "six sigma" famously equals 3.4 defects per million rather than the roughly two-per-billion that a perfectly centered six-sigma process would give in pure statistics. The 3.4 figure is really the long-term defect rate of a process that is six sigma in the short term but allowed to drift 1.5 sigma, which in statistical terms is equivalent to a centered 4.5-sigma process. The distinction has a name: short-term sigma describes how the process performs right now in a controlled snapshot; long-term sigma describes how it performs over months of real conditions. The shift is the bridge between the two.
By the numbers. A six-sigma level of performance corresponds to 3.4 defects per million opportunities, a benchmark that assumes the 1.5-sigma long-term shift first proposed at Motorola. Standard conversion tables give roughly 66,800 DPMO at three sigma, about 6,210 at four sigma, and about 233 at five sigma. See MoreSteam's Six Sigma conversion table and iSixSigma's definition of the Six Sigma metric.
What sigma level should you aim for?
The honest answer is that it depends on the consequence of a defect, not on a universal target. Three sigma, around 66,800 defects per million, sounds respectable until you translate it: in a high-volume line or a safety-critical part, a 6.7% defect rate is a fire. Four sigma is a common, defensible baseline for many manufactured characteristics, and it already demands real process discipline. Five and six sigma are worth the effort only where a single escape is catastrophic, a medical device, an aerospace fastener, a food-safety control, or where volumes are so large that even 233 defects per million adds up to a costly recall. The trap is treating six sigma as a moral goal for everything; the discipline is matching the target to what a defect actually costs, then investing there and nowhere else.
How do you estimate a process's sigma level?
You estimate a sigma level by counting defects against opportunities, converting to a defect rate, and reading the corresponding sigma from the table. The arithmetic is deliberately simple; the judgment is in defining "defect" and "opportunity" honestly.
- Define a defect and an opportunity. A defect is any failure to meet a spec; an opportunity is each chance to create one. Loose definitions inflate your sigma level, so pin these down before counting.
- Count over a real window. Tally the defects and the total opportunities across enough production to be representative, a single good shift is not a sigma level.
- Compute DPMO. Divide total defects by total opportunities, then multiply by one million. This is defects per million opportunities the input to the table.
- Convert DPMO to a sigma level. Read across the standard table (or use the conversion formula). 6,210 DPMO is four sigma; 66,800 is three.
- Decide whether you have short-term or long-term data. A short controlled study gives short-term sigma; months of routine production give long-term. Know which one you are quoting before you compare it to anyone else's.
- Confirm the process is stable first. A sigma level from an out-of-control process is meaningless, the distribution is not stable, so the tail math does not hold. Check stability on a control chart before you trust the number.
How should you use sigma level, and how not to?
Use sigma level as a comparable, honest snapshot of capability, and resist the temptation to treat it as a scoreboard to be gamed. The number is genuinely useful for prioritizing: a characteristic running at three sigma is bleeding defects and deserves attention, while one at five sigma probably does not repay further investment. It also travels well, a sigma level on Line 2 in one plant means the same thing as a sigma level on Line 7 in another, which raw defect counts never do.
The misuses are just as consistent. Chasing six sigma on every characteristic burns money where four sigma would be plenty. Quoting a short-term sigma from a cherry-picked study and comparing it to someone's long-term number is apples to oranges. And computing a sigma level on an unstable process, or on defect definitions loose enough to flatter the result, produces a confident number that means nothing. The tool sits inside the broader discipline of statistical process control and the seven basic quality tools and like all of them it is only as honest as the data feeding it.
That data honesty is the quiet prerequisite. A sigma level calculated from defects logged late, categorized loosely, or missed entirely at shift end describes a plant that does not exist. When defects and opportunities are captured in real time at the line, the sigma level reflects what actually happened rather than what someone reconstructed, the same data foundation CLS built in moving off paper. Define your terms, confirm stability, know whether you are quoting short- or long-term, and the sigma level becomes one of the cleanest ways to say how good a process really is. For the broader lean context, see lean manufacturing.