Three dependent variables, derived from an extended Signal-Detection paradigma, were used in each of 3 experiments: memory performance, confidence-level, and response-bias. Each memory item was first judged by S and then fictitiously by 2 confederates providing different degrees of agreement and disagreement. As compared to agreement moderate disagreement yielded both better recognition performance and, if S's judgements were false, less confidence. Strong disagreement failed to repeat these findings. Balanced agreement/disagreement raised the level of both performance and confidence relative to a situation without information from the group. In all the experiments correct decisions yielded higher confidence than errors. Festinger's theory of social comparison processes accounts for all results in performance, but for explaining the confidence shifts assumptions on ‘internal cues’ should also be incorporated. The response-bias was not affected by social treatment differences, thus supporting the view of some Signal-Detection theorists. Proposals towards a general theory of stimulus processing in social context are outlined and some of its consequences are discussed by taking as examples some conformity experiments.