DPMO and sigma level: the formula, the table, and a worked example.
DPMO — defects per million opportunities — is how Six Sigma puts a number on quality so a stamping line and a software process can be compared on one scale. This explains the formula, converts DPMO to a sigma level, and works a real example end to end.
A 2% reject rate sounds bad on a press shop and fine on an assembly line, but those words mean nothing across processes. DPMO replaces the adjectives with a count: how many defects you would expect per million chances to make one. From that single number you read a sigma level, and a "3.2 sigma process" tells everyone the same thing whether they run machines or write code.
What DPMO actually measures
Three terms have to be clear before the formula makes sense:
- Defect: any failure to meet a requirement. One part can carry several defects.
- Defective: a unit with one or more defects. This is the difference people miss — 50 defects might sit on 30 defective units.
- Opportunity: each independent chance to create a defect on a unit. A bracket checked on 5 characteristics has 5 opportunities.
DPMO scales the defect count by both the number of units and the opportunities per unit. That is why it travels across processes: a part with 5 inspection points and a form with 20 fields are put on the same per-opportunity footing.
The DPMO formula
The formula is:
A related figure is DPO (defects per opportunity), which is the same thing before scaling: DPO = DPMO ÷ 1,000,000. And yield at the opportunity level is simply 1 − DPO, expressed as a percentage.
Worked example: a machined bracket line
Take a month on a CNC bracket line:
- Units produced: 1,200 brackets
- Opportunities per unit: 8 inspected characteristics
- Defects found: 19 (across all units and all characteristics)
Total opportunities = 1,200 × 8 = 9,600. Then:
DPMO = (19 ÷ 9,600) × 1,000,000 = 1,979 DPMO.
Reading the conversion table below, ~1,979 DPMO sits just above the 4.5 sigma line (which is 1,350 DPMO) and below 4.0 sigma (6,210 DPMO), so this line runs at roughly 4.3 sigma. The opportunity-level yield is 1 − (1,979 ÷ 1,000,000) = 99.80%. The DPMO and sigma level calculator does this conversion both ways — enter defects, units and opportunities to get DPMO and sigma, or enter a target sigma to see the DPMO you must hit.
DPMO to sigma level conversion table
These values include the standard 1.5 sigma shift used in Six Sigma practice (explained next), so they are "process sigma" figures.
| Sigma level | DPMO | Yield |
|---|---|---|
| 1.0 | 691,462 | 30.9% |
| 2.0 | 308,538 | 69.1% |
| 3.0 | 66,807 | 93.3% |
| 3.5 | 22,750 | 97.7% |
| 4.0 | 6,210 | 99.38% |
| 4.5 | 1,350 | 99.87% |
| 5.0 | 233 | 99.977% |
| 5.5 | 32 | 99.997% |
| 6.0 | 3.4 | 99.99966% |
The famous headline — "Six Sigma means 3.4 defects per million" — is the bottom row. Note how steep the climb is at the top: moving from 5 to 6 sigma cuts DPMO from 233 to 3.4, nearly 70-fold, which is why the last sigma is by far the hardest and most expensive to win.
The 1.5 sigma shift, explained
Here is the part that confuses people. A true ±6 sigma spread under a perfectly centred normal curve gives about 2 defects per billion, not 3.4 per million. The 3.4 figure comes from deliberately allowing the process mean to drift by 1.5 sigma over the long term.
The reasoning, from Motorola's original work, is that no process stays perfectly centred forever — tool wear, material lots, shift changes and temperature all nudge the mean. So Six Sigma builds in a 1.5 sigma long-term drift as a safety margin. The result: a process that is 6 sigma in the short term is treated as 4.5 sigma long-term, and 4.5 sigma centred is the 3.4 DPMO you see quoted.
How DPMO relates to Cp and Cpk
DPMO and process capability describe the same reality from two directions. Capability indices like Cp and Cpk work from the spread and centring of measured data; DPMO works from counted defects. They line up: a centred process with Cpk = 1.0 corresponds to about 3 sigma (66,807 DPMO), Cpk = 1.33 to about 4 sigma, and Cpk = 1.5 to about 4.5 sigma — the common automotive PPAP target.
Use whichever you can measure. For dimensional features with variable data, Cpk from the process capability method is richer because it sees the distribution. For attribute data — pass/fail, present/absent, paperwork errors — you cannot compute Cpk, so DPMO is the right tool.
Common mistakes
- Confusing defects with defectives. Three defects on one unit is three defects but one defective. DPMO uses the defect count; first-pass yield uses defectives.
- Inflating opportunities. Counting every conceivable feature as an opportunity quietly lowers your DPMO and flatters the sigma. Fix the opportunity definition once and keep it stable, or trends become meaningless.
- Forgetting the 1.5 shift. Comparing a short-term sigma to a long-term target overstates performance by 1.5 sigma.
- Reading DPMO as a percentage. 6,210 DPMO is 0.62%, not 6.2%. The "per million" base trips people up.
DPMO tells you how often a process fails; it does not tell you where the defects come from. Pair it with the Cp/Cpk guide for variable data, the AQL sampling guide for inspection planning, and a way to catch defects at source — CadNexa's auto-ballooning tool reads a PDF drawing with Smart Detect and Box+Balloon OCR and builds the inspection sheet that feeds your defect counts. Ready-made quality templates round out the workflow.
Frequently asked questions
What is a good DPMO?
It depends on the industry, but as a rule of thumb: above 66,807 DPMO (below 3 sigma) is poor, around 6,210 DPMO (4 sigma) is typical for many manufacturing lines, and 3.4 DPMO (6 sigma) is world-class. Automotive PPAP work generally targets 4.5 sigma or better, which is 1,350 DPMO.
How do you calculate sigma level from DPMO?
Convert DPMO to a defect rate per opportunity (DPMO / 1,000,000), find the corresponding z-value of the normal distribution for that tail area, then add the 1.5 shift to express it as process sigma. In practice you read it from a DPMO-to-sigma table or a calculator rather than computing the z-value by hand.
Why is Six Sigma 3.4 defects per million and not 2 per billion?
A perfectly centred 6 sigma process gives about 2 defects per billion. Six Sigma deliberately allows a 1.5 sigma long-term drift of the process mean to account for real-world variation, which moves the figure to 3.4 defects per million — the long-term, customer-experienced rate.
What is the difference between DPMO and PPM?
PPM (parts per million) counts defective units per million units, ignoring how many opportunities each unit has. DPMO counts defects per million opportunities, normalising for product complexity. DPMO lets you compare a simple part and a complex one fairly; PPM does not.