AQL Sampling vs. 100% Inspection: What Each One Misses

What is AQL sampling?

AQL sampling is a lot-acceptance method: you inspect a random sample, count defects, and accept or reject the whole lot against a preset limit. AQL stands for acceptance quality limit (often called acceptable quality level), the worst average defect rate a buyer will tolerate.

The tables come from MIL-STD-105, which the U.S. Department of Defense issued in 1950 and withdrew in 1995. Its civilian successors, ANSI/ASQ Z1.4 and ISO 2859-1, reproduce the same tables and are still the default in most supplier agreements.

How a plan works, in four steps:

1. Find your lot size and pick an inspection level. General Level II is the usual default.
2. Read the code letter from the table. A lot of 1,201 to 3,200 units at Level II is code K.
3. Read the sample size and limits for your AQL. Code K at AQL 1.0 means inspect 125 units, accept the lot at 3 or fewer defects, reject at 4 or more.
4. Accept or reject the whole lot based on that count.

The appeal is real. Sampling is cheap, fast and contractually familiar, and for stable processes making discrete, interchangeable units, it does its job.

What AQL sampling misses

AQL sampling misses every defect outside the sample, by design. It answers "is this lot probably acceptable?", not "where are the defects?" On a continuous line, that gap gets wider for four reasons.

It accepts bad lots more often than most people assume. Take the plan above: 125 samples, accept at 3. A lot that is 1% defective passes 96% of the time. A lot that is 2% defective still passes 76% of the time, and one that is 3% defective passes about half the time. The AQL is a ceiling on average quality, not a target.

It assumes random, independent units. The math requires samples drawn at random from a uniform lot. A roll of film or a coil of painted steel is not a bag of interchangeable parts. Defects on a web cluster: a contaminated roller, a coating streak or a humidity swing can mark hundreds of feet in a row, then stop.

It samples where it is easy, not where it is random. On a roll, the practical sample is the outer wrap or a cut at the core. A burst of pinholes in the middle of the roll is invisible to both.

It tells you nothing about why. A pass or fail per lot carries no position, no timestamp and no line conditions. When a defect escapes, the record cannot tell you when it started or what changed.

What manual 100% inspection misses

Manual 100% inspection misses about 1 in 5 defects, even with skilled people. A Sandia National Laboratories study of 82 trained inspectors checking 140 precision parts found they caught 85% of defective items and wrongly rejected 35% of good ones. The authors put the industry average catch rate at about 80%.

On a moving line, the limits get tighter:

• Speed. A web at hundreds of feet per minute is faster than any person can scan in detail. Lines get slowed, or the inspection becomes a glance.
• Fatigue and repetition. Attention drops across a shift, and the rare defect is the easiest to miss.
• Reflective and clear substrates. Pinholes in clear film or flaws on a mirror finish are hard to see under shop lighting.
• Overcalls. When in doubt, people reject. That 35% false-reject rate in the Sandia study is good product turned into scrap.
• Cost. It takes a person, or several, on every line, every shift.

And like sampling, a manual check rarely leaves a record of where on the roll a defect sat, so the root cause still goes unfound.

What automated 100% inspection changes

Automated 100% inspection closes both gaps: it sees every inch at full line speed, and it remembers where every defect was. That second part is what turns inspection into production intelligence.

Here is what that looks like on real lines Shelfmark runs:

• A football-field-long webbing roll at the largest U.S. parachute manufacturer used to be unrolled and checked by hand, foot by foot. Shelfmark now inspects the whole roll in six minutes and tags every defect. (case study)
• Mirror-finish industrial film is inspected at 400 feet per minute, a surface that throws light straight back at a person or a standard camera. (case study)
• A decorated apparel line where about 10% of prints needed rework and every product was checked by hand cut print defects by more than 90%. (case study)

Because every defect is logged with its position on the roll and a timestamp, you can cut out the bad section instead of downgrading the whole roll. Add line data such as temperature and humidity, and patterns appear. On one line, defect rates rose and fell with the weather. The fix was a storm door. (case study)

The old objection to automated inspection was the work: tuning rules, labeling images, retraining models for every new SKU. With a managed service, that work sits with the vendor, not your quality team.

AQL vs. 100% inspection at a glance

When AQL still fits: incoming inspection of discrete parts from a stable supplier, or a contract that requires it. Many plants keep AQL for the paperwork and add 100% inspection on the line.

When to move past it: an escaped defect costs you a chargeback, a hold or a customer; defects come in bursts; or your buyer wants proof of quality for every run, not a pass on a sample.

See what your sample is missing

In 30 minutes, a Shelfmark expert can tell you whether we can catch your defects on your substrate at your line speed. No capital request: Shelfmark runs as a monthly subscription that covers the cameras, the platform and the team managing it.

Book a free line fit assessment →

Frequently asked questions

Keep reading