Defect Classification: Critical, Major, Minor

Defect classification sorts every possible defect into three severity classes, critical, major, and minor, based on how much it hurts the user or the product. Critical defects threaten safety or make the product fail; major defects impair function or usability; minor defects are cosmetic and do not affect use. The class you assign drives how tightly you sample and inspect for it.

Classification is the quiet decision that everything else in acceptance sampling hangs on. Set the classes well and your inspectors spend their attention where it matters and your accept/reject calls are defensible. Set them badly, or leave them to each inspector's judgment, and you either ship dangerous product or scrap good product, sometimes both in the same week. This guide defines the three classes, connects them to AQL sampling, and shows how to build a classification standard your inspectors actually apply the same way.

What are the three defect classes?

The three-way split, critical, major, minor, comes from acceptance-sampling practice codified in standards like ANSI/ASQ Z1.4 and its predecessor MIL-STD-105. Each class is defined by consequence, not by how obvious the defect is or how much it annoys anyone.

The line that trips people up is critical versus major. The test is safety and legality: if the defect could hurt someone or break a regulation, it is critical, no matter how small. A hairline crack you can barely see in a bracket that holds a person's weight is critical. A cracked-off corner on a decorative trim panel is minor. The defect's appearance tells you nothing about its class; its consequence tells you everything.

Why does defect classification drive AQL sampling?

Because you set a different Acceptance Quality Limit for each class, and the AQL decides how many defects a sampled lot can contain before you reject it. A tighter AQL means a smaller allowed fraction of defects, which means a stricter accept/reject rule. Critical defects get the tightest AQL, commonly zero, so a single critical defect in the sample rejects the whole lot. Major defects get a middle AQL (often around 1.0-2.5), and minor defects get a looser one (often 2.5-4.0).

This is the whole reason classification exists operationally: it lets you spend inspection strictness in proportion to risk. You are not equally worried about a scratch and a live wire, so you do not apply the same acceptance rule to both. On a single inspection of one lot you run three parallel accept/reject decisions at once, one per class, each against its own AQL and allowable-defect count. Get the classification right and the AQL system automatically applies the right pressure; get it wrong and you either wave through hazards or reject lots over cosmetics. The mechanics of pulling the sample size and acceptance number from the tables live in the AQL guide.

How many defect classes should you use?

Three is the standard, and for most operations three is enough. But some industries split the middle. It is common to see majors divided into "major A" and "major B," where major A sits just below critical (a functional failure with cost or warranty exposure) and major B is a lesser functional issue, each with its own AQL. Highly regulated or high-liability products sometimes add finer gradations still. The trap is going the other way and using too few: collapsing everything into "pass or fail" throws away the whole point, which is to spend inspection strictness in proportion to consequence.

The right number of classes is the smallest number that lets you set a defensibly different acceptance rule for each. If two classes would always carry the same AQL and the same reaction plan, they are one class wearing two names. If a group of defects clearly deserves a tighter gate than the rest of its class, it has earned its own class. Do not add classes for tidiness; add them only when the accept/reject decision genuinely differs. And whatever number you land on, the customer contract or regulation can override you, some customers mandate their own classification scheme, and when they do, theirs wins.

How do you build a defect classification standard?

The goal is that any two inspectors, on any two shifts, put the same defect in the same class. That only happens when the classification lives in a written, illustrated standard, not in people's heads. Build it like this.

1. List real defects, not categories. Start from your actual reject history and complaints. Name specific, observed defects (chip on sealing face, missing label, bent pin), not vague buckets like "surface issues."
2. Classify each by consequence. Run every listed defect through the two-question test, safety/regulation first, then function, to assign critical, major, or minor. Involve engineering and quality so the safety calls are defensible.
3. Show the boundary cases with photos. The disagreements always cluster at the borders. A limit sample, a photo of the worst acceptable and the best rejectable, settles more arguments than a paragraph ever will.
4. Assign an AQL to each class. Set the acceptance limit per class, honoring any customer or regulatory requirement. Critical is normally zero.
5. Tie it to the reaction plan. Define what happens when each class is found: a critical finding stops the lot and escalates, a major triggers a hold and disposition, a minor is logged and trended.
6. Train, then check agreement. Have several inspectors classify the same defect set and compare. Where they split, the standard is unclear, fix the standard, not the inspector.
7. Review it against reality quarterly. A defect that keeps escaping to customers may be misclassified, or the class may be right and the process wrong. New failure modes get added; the standard is living.

What happens when classification goes wrong?

Two failure modes, opposite symptoms. Over-classification treats cosmetic blemishes as majors, so good lots get rejected, scrap and rework climb, and inspectors quietly start ignoring the standard because it cries wolf. Under-classification treats a functional or safety defect as minor, so hazardous product sails through the loose minor gate and reaches customers, the failure nobody sees until a complaint or a recall. Inconsistent classification does both at random, which is worse, because now the data is noise and you cannot even see which direction you are failing.

The tell is inspector disagreement. When the same defect gets called major by days and minor by nights, the standard is not doing its job, and the reject data becomes untrustworthy for every downstream decision, from supplier scorecards to defect tracking to AQL sampling itself. Classification is the input; if it is noisy, every number built on it is noisy too.

Making classification consistent on the floor

A classification standard only works if inspectors can reach it in the second they need it, at the station, mid-inspection, not in a binder across the plant. When the defect list, the limit-sample photos, and the reaction plan live where the inspection happens, and the call itself is captured digitally with the defect, classification stops drifting between shifts and the data downstream finally means something. That is the kind of connected-worker quality guidance Harmony's platform is built to put at the point of work, on top of your existing process with no rip-and-replace.

Good classification also feeds the rest of the quality system cleanly. A critical or major finding routes into your control-of-nonconforming-product process for hold and disposition, feeds a non-conformance report and sharpens where incoming material inspection and attribute inspection spend their attention. In batch production where a single classified critical defect can quarantine an entire lot, getting the class right is the difference between holding one bad batch and shipping a recall. You can see the full quality loop in our CLS case study.