Two experiments were conducted to evaluate the context switch effect upon retrieval of the information about a cue-outcome relationship in human predictive learning. The results replicated the well-known effect of renewal of the cue-outcome relationship due to a context change after a retroactive interference treatment, as much as the null effect of the context change upon acquisition before retroactive interference training had taken place (Experiment 2). However, retrieval of an unambiguous cue-outcome relationship was also impaired by a context switch when this relationship was established in a context where a different cue had received an interference treatment (Experiments 1 and 2). Once the interference treatment was given to participants in one context, unambiguous cue-outcome relationships learned in a different context also became context specific (Experiment 2). The implications of these results for retrieval theory are discussed. (Contains 3 tables and 2 figures.)