Subjects in this study were asked to infer an attribute of a target person on the basis of his report about himself: Two informational determinants of such inferences were varied: (a) reliability of the report, i.e. the belief that the target person would correctly report events that had actually occurred and (b) diagnosticity of the actual events, i.e. the belief that the actual events are indicative of the attribute. Normative considerations require that the effect of diagnosticity be dependent on reliability so that as reliability increases, judgment should become less regressive, i.e. vary more as a function of events' diagnosticity. The results indicate that subjects employed a simple but inappropriate averaging rule in combining reliability and diagnosticity information. This rule, like many other simplifying judgmental heuristics, resulted in inferences that were more extreme than warranted by normative models. The inappropriate combination of reliability and diagnosticity information may thus contribute to observers' tendency to over-attribute personal characteristics to others.