When anxious or depressed people try to recall emotionally ambiguous events, they produce errors that reflect their habits of interpreting ambiguity in negative ways. These distortions are revealed by experiments that evaluate performance on memory tasks after taking interpretation biases into account—an alternative to the standard memory-bias procedure that examines the accuracy of memory for clearly emotional material. To help establish the causal role of interpretation bias in generating memory bias, these disortions have been simulated by training interpretation biases in nondisordered groups. The practical implications of these findings for therapeutic intervention are discussed; future directions are described.