The aim of the current study was to examine, through posthoc analyses, changes in and predictive utility of mood‐related cognitive variables as a function of treatment modality in a group of alcohol dependent patients with elevated depressive symptoms. In addition to the background partial hospital treatment for alcoholism which lasted a mean of 21.2 days, study patients (n = 35) received cognitive‐behavioral treatment for depression (CBT‐D) or a control treatment consisting of relaxation training (RTC). While both groups showed improvement on dysfunctional attitudes during treatment, only the CBT‐D group improved on measures of alcohol‐related expectancies. Changes in cognitive variables during treatment predicted drinking outcomes, and the predictors of drinking outcomes varied across the two treatments. Among the RTC patients, changes in positive alcohol‐related expectancies were negatively correlated with drinking frequency and quantity at follow‐up. However, for the CBT‐D patients, changes in self‐efficacy concerning negative mood situations and negative alcohol‐related expectancies were negatively correlated with drinking at longer‐term follow‐up. The results of this study provide evidence concerning the mechanisms by which the treatment modalities examined may affect patient outcome. Although these results are preliminary in nature, they do suggest that future research might examine efforts to capitalize on these mechanisms through the facilitation of changes found to predict better drinking outcomes in this study.