Prevailing models of exposure therapy for phobias and anxiety disorders construe level of fear throughout exposure trials as an index of corrective learning. However, the evidence, reviewed herein, indicates that neither the degree by which fear reduces nor the ending fear level predict therapeutic outcome. Developments in the theory and science of fear extinction, and learning and memory, indicate that 'performance during training' is not commensurate with learning at the process level. Inhibitory learning is recognized as being central to extinction and access to secondary inhibitory associations is subject to influences such as context and time, rather than fear during extinction training. Strategies for enhancing inhibitory learning, and its retrieval over time and context, are reviewed along with their clinical implications for exposure therapy and directions for future research.