Visual inspection criteria in manufacturing usually say what to look for. They rarely show where a pass turns into a reject, and that is exactly where two inspectors disagree.

Joao Passed It. Sunil Put It in the Rework Bin.

Joao passed the part on first shift. On second shift, Sunil put the same part in the rework bin.

Both followed the same sheet. It said “no excessive scratches on the sealing face.” Neither was careless. Each had a different picture of “excessive,” and the sheet had no picture at all.

The plant paid twice. Once for the rework Sunil triggered, and once for the unknown number of parts Joao passed that Sunil would have caught.

What Are Visual Inspection Criteria?

Visual inspection criteria are the written rules that tell an inspector when a part passes, fails or needs a second opinion. A good set names the defect, where on the part it matters, how big is too big, and what to do when it’s unclear.

In my experience, most sets name the defect and stop there. “No burrs.” “No excessive scratches.” “No visible damage.” Those are descriptions of a feeling, not a decision rule.

This Isn’t a Joao and Sunil Problem

Inspectors disagree because human inspection varies, and no sheet removes that on its own. The research on this is older than most plants and it keeps finding the same thing.

A 2014 presentation by Judi See of Sandia National Laboratories on a visual inspection reliability study summarizes the earlier research this way: “Error rates of 20% to 30% are common.” “Large individual differences exist.” “No defect is detected by all inspectors.” Treat it with care. Those lines come from the deck’s literature review of inspection research, not from the study’s own results, which the deck labels preliminary. It’s a slide presentation made for a federal interagency manufacturing group, and it isn’t about your line. Even so, it matches what I’ve seen on floors: the disagreement is predictable, and it sits on the borderline parts.

The same presentation lists training, inspection procedures and apparatus as the main ways to improve inspection, and notes that a review of the organization’s own specification exception releases found procedural problems, including “misinterpretation or lack of clarity.” In plain terms: part of the variation lives in the document.

What Happens When Inspectors Meet a Borderline Part?

They split, and the split is measurable. A technical article published by ISPE in January/February 2020, written by Sambhujyoti Das, describes an evaluation of three visual inspectors on 100 vials, with three independent trials on three different days. It’s a pharmaceutical setting, so read it as an illustration of the pattern, not a benchmark for your parts. One vial was misclassified “88.9% times across all inspectors,” which the author calls “indicative of a borderline case.” The author’s recommendation was to re-inspect it and replace it in future trials with a clearly good vial.

That is a fair fix for a test. On a real line you can’t replace the vial. The borderline part will keep arriving, so the decision about it has to be written down once, with a picture, and handed to every shift. This is the same shift-to-shift gap we covered in operator-to-operator variation, applied to the inspection bench.

Why Does the Gap Turn into Rework?

Because the inspector who is stricter sends parts back, and the one who is looser passes them on. Both feel correct.

That’s how rework appears with nobody making a mistake. If you want the cost side, our piece on how to measure rework in manufacturing shows how to count it, and the article on rework instructions covers what to do once a part is in the bin. This piece is upstream of both: how a part ends up there.

What the Market Sells Instead

Mostly counting. Quality modules that log rejects by category and plot them by week. Vision systems that flag defects after someone has labeled what a defect is.

Both help, and neither answers the question on Joao’s bench. A chart of rejects tells you Sunil’s shift sends back more parts. An algorithm trained on labeled images inherits whoever did the labeling. In my view the labeling is the unsolved part, and it starts with an inspector being able to say where the line is.

How to Write Visual Inspection Criteria That Inspectors Agree On

Here is the five-part criterion I’d use for any defect that causes disagreement. It’s my own practice, not a published standard.

1. Name the defect. One word the whole plant uses.
2. Name the zone. A scratch on a sealing face matters; the same scratch on a non-functional edge may not.
3. Set the limit in something you can measure or compare. Length, count, depth by fingernail catch, or “no larger than the reference sample.”
4. Show three photos. A clear pass, a borderline that passes, a borderline that fails, each with the decision written beside it.
5. Say who decides when it’s unclear. A named role, with a time limit, so the part doesn’t sit.

The photos are the part most sheets skip, and the part that carries the information. Our article on how to standardize work across shifts covers how to keep the sheet identical on every shift once it exists.

How We Capture the Inspector’s Eye

At SenseiLab, we capture tacit knowledge directly from the person who has it. For inspection, that means sitting with the senior inspector at the bench and having them work through real parts: where they look first, what a pass looks like next to a reject, and which part always splits the room.

The senior inspector describes it in their own words. The supervisor validates it before it reaches the floor, and the borderline calls become part of the procedure. The eye stops being a private asset. Our guide to documenting tribal knowledge in a plant explains the capture method in more depth.

Run the Five-Part Boundary Test This Week

You can do this without us.

1. Pull five parts: two clear passes, two clear rejects, one borderline.
2. Hand them to each inspector separately. Silent calls only.
3. Record each call, and count the disagreements.
4. Put the borderline part on the table with both inspectors and decide together.
5. Photograph it, write the decision beside the photo, and attach it to the work instruction by Friday.

In my experience the clear parts get matching calls and the borderline one doesn’t. That one part is your first criterion, and the rest of the work is repeating the exercise for the next defect.

Frequently Asked Questions About Visual Inspection Criteria

What are visual inspection criteria in manufacturing?

Visual inspection criteria are the written rules that tell an inspector when a part passes, fails or needs a second opinion. They work best when they name the defect and the zone, set a limit, and show photos of the line between pass and reject.

Why do inspectors disagree on the same part?

Because human inspection varies and borderline parts invite different judgments. A 2014 Sandia presentation on visual inspection reliability summarized earlier research: large individual differences exist, and no defect is detected by all inspectors.

How do we reduce rework or defects caused by inconsistent processes?

Find where two people doing the same job reach different results, and write down the decision rule with a picture. For the full method, see our article on how to reduce defects caused by inconsistent processes.

How many reference photos does a work instruction need?

For a defect that causes disagreement, three: a clear pass, a borderline that passes, and a borderline that fails. That is my own practice, not a standard, and you should add more if inspectors still split.

Forward this to one quality lead who suspects their shifts don’t agree on what a reject looks like.

— Diego Echenique, SenseiLab.