Abstract Misclassification of categorical variables creates problems for analysis and interpretation, leading to biased estimates if the misclassification is ignored. For example, in two‐way tables, its effect is often to bias observed associations toward the null value. In three‐way tables, an association between two variables can be biased in any direction when a confounder variable is misclassified. More generally, complex combinations of effects can occur, and understanding them requires information about the misclassification parameters. Such information can be obtained from validation studies or repeated measurements. These also allow the misclassification biases to be removed or reduced using appropriately modified estimators. The most straightforward adjustment methods are based on simple back‐calculation, while in more complex cases, formal modeling methods are more suitable. These are essentially models for contingency tables where some variables are partially or completely unobserved.