Attribute sampling is binary and rigid – either an attribute is present in your test or it isn’t. For example, there should never be any traces of melamine or poison found in sample tests from cat food. A test revealing even a small amount would represent a severe quality problem requiring immediate action. But there are situations in which attribute sampling can get tricky. For example, you wouldn’t want anyone slipping arsenic into your water. But it’s a naturally occurring element, and it’s likely that it would show up in trace amounts in quality tests of tap water. In a case like this, you would need to do variance sampling, because attribute sampling of arsenic in water is inevitably going to be positive.

Variance sampling is less rigid – a certain level of imperfection is tolerated. In the case of arsenic, you would test for a specific number of parts per million that is considered safe. Or suppose your customer’s specifications for high-quality dog food specifies primarily choice beef, but a certain amount of gristle is allowed. You might design tests that would capture the percentage of gristle, and 2 percent, for example might represent an allowable variance from 100 percent pure meat. Per your client’s specifications, this would be an allowable variance, agreed upon up front.