The attributional vulnerability model of depression has rarely been tested in prospective designs, or as an interaction of stressful events and cognitions, or with depression as a specific response outcome. Moreover, the model has rarely been applied to children. All these issues were addressed in this study of prediction of diagnoses during a six‐month follow‐up for a sample of children that included offspring of women with affective disorders presumed to be at high risk for depression. Hierarchical multiple regression analyses failed to support the attributional hypotheses: depression was best predicted by initial symptoms and by life stress but not by attributions for negative outcomes or the interaction of attributions and life‐events. Non‐affective diagnoses, on the other hand, were predicted by an interaction of life‐events and attributions. The results suggest limitations in the range of application of the attributional model in clinical samples, at least with children and adolescent