This study extended that of Kwon and Oei [Kwon, S.M., Oei, T.P.S., 2003. Cognitive change processes in a group cognitive behavior therapy of depression. J. Behav. Ther. Exp. Psychiatry, 3, 73-85], which outlined a number of testable models based on Beck's cognitive theory of depression. Specifically, the current study tested the following four competing models: the causal, consequential, fully and partially interactive cognitive models in patients with major depressive disorder.A total of 168 clinically depressed outpatients were recruited into a 12-week group cognitive behaviour therapy program. Data was collected at three time points: baseline, mid- and at termination of therapy using the ATQ, DAS and BDI. The data were analysed with Amos 4.01(Arbuckle, J.L., 1999. Amos 4.1. Smallwaters, Chicago.) structural equation modelling.Results indicated that dysfunctional attitudes, negative automatic thoughts and symptoms of depression reduced significantly during treatment. Both the causal and consequential models equally provided an adequate fit to the data. The fully interactive model provided the best fit. However, after removing non-significant pathways, it was found that reduced depressive symptom contributed to reduced depressogenic automatic thoughts and dysfunctional attitudes, not the reverse.These findings did not fully support Beck's cognitive theory of depression that cognitions are primary in the reduction of depressed mood.