This study examined the cognitive mediation of relapse prevention by cognitive therapy (CT) in a trial of 158 patients with residual depression. Scores based on agreement with item content of 5 questionnaires of depression-related cognition provided no evidence for cognitive mediation. A measure of the form of response to those questionnaires, the number of times patients used extreme response categories ("totally agree" and "totally disagree"), showed significant and substantial prediction of relapse, differential response to CT. and conformity to mediational criteria. CT reduced relapse through reductions in absolutist, dichotomous thinking style. CT may prevent relapse by training patients to change the way that they process depression-related material rather than by changing belief in depressive thought content.