1. Memory is assessed by measuring retrieval which is often elicited by the solely presentation of the conditioned stimulus (CS). However, as known since Pavlov, presentation of the CS alone generates extinction. 2. One-trial avoidance (IA) is a much used conditioned fear paradigm in which the CS is the safe part of a training apparatus, the unconditioned stimulus (US) is a footshock and the conditioned response (CR) is to stay in the safe area. Retrieval of the memory for the step-down version of this task is measured in the absence of the US, as latency to step-down from the safe area (i.e., a platform). 3. Extinction of the IA response is installed at the moment of the first non-reinforced test session, as clearly shown by the fact that many drugs, including PKA, ERK and protein synthesis inhibitors as well as NMDA receptor antagonists, hinder extinction when infused into the hippocampus or the basolateral amygdala at the moment of the first test session but not later. 4. Some, but not all the molecular systems required for extinction are also activated by retrieval, further endorsing the hypothesis that although retrieval is necessary for the generation of extinction this last process constitutes a new learning secondary to the non-reinforced expression of the original trace.