Retrieval procedures, if carried out without reinforcement, initiate memory extinction. The extinction of onetrial avoidance learning requires glutamate NMDA receptors, calcium-calmodulin dependent protein kinase II, cAMPdependent protein kinase and mitogen-activated protein kinases, and, importantly, protein synthesis and gene expression in the hippocampus. The extinction of fear-potentiated startle requires NMDA receptors and mitogen-activated protein kinases in the basolateral amygdala. The extinction of conditioned taste aversion requires protein synthesis in the insular cortex. Thus, extinction is an active process that involves a variety of molecular events and -at least for the one task in which it was studied- both gene expression and protein synthesis. Insofar as in each of the tasks mentioned, the treatments were studied in a different brain region, therefore, it is still not known whether extinction in general uses those brain areas in parallel, or whether the extinction of each task is metabolically different. A role of endogenous cannabinoids in extinction has been postulated, some evidence indicate that they act on the amygdala, but indirect findings suggest that they may also exert their action in the hippocampus. When carried out using methods that enhance perception that the reinforcement is absent, extinction can be quite profound, and the animals require “de novo” gene expression and protein synthesis in the hippocampus in order to reinstall the original learning. This might be of value in the design of “exposure” therapies for the treatment of phobias and of post-traumatic stress disorders.