Five experiments are reported showing that the interpretation of personally relevant emotional information can be modified by systematic exposure to congruent exemplars. Participants were induced to interpret ambiguous information in a relatively threatening or a benign way. Comparison with a baseline condition suggested that negative and positive induction had similar but opposing effects. Induction of an interpretative bias did not require active generation of personally relevant meanings, but such active processing was necessary before state anxiety changed in parallel with the induced interpretative bias. These findings provide evidence consistent with a causal link between the deployment of interpretative bias and anxiety and reveal something of the processes underlying this association.