This study demonstrates how mixture distribution item response models can be used to detect different response styles in the clinical assessment of anger expression. Analyses of 3 subscales of the State-Trait Anger Expression Inventory in a clinical sample of 4,497 patients revealed that there are different response styles that manifest themselves in 2- and 3-class solutions. These solutions are robust across subsamples. Response styles reflect both psychologically meaningful biases (i.e., social desirability) and nonmeaningful response category preferences. Person parameters that correct for class membership (and thus, for response styles) are computed and compared with raw scores. The implications of these results for research on clinical assessment are discussed.