Two experiments explored the role of irrelevant instructions on performance after reversal of conditional discriminations in human beings. Students were exposed to a matching-to-sample task where they had to choose between two comparative stimuli based on the form of a sample stimulus. During Phase 2 the criterion was reversed, and participants had to make their choice based on the color of the stimulus. The frequency of responses to form was recorded in the absence of feedback during the test. Experiment 1 manipulated the instructional context where each phase took place. Two different instructions irrelevant to the task were counterbalanced as instructions 1 and 2. Group 111 received the training with instruction 1 present. Group 112 was tested in the presence of instruction 2. Finally, Group 121 received training in the presence of instruction 1, reversal in the presence of instruction 2, and testing in the presence of instruction 1. Instructions change between reversal training and testing renewed Phase 1 performance, particularly when the test was conducted in the presence of instruction 1 (Group 121). Renewal disappeared when the semantic meaning of the instruction was eliminated (Experiment 2). Instructions may be used as physical contexts in human conditional discrimination reversal. Key words: renewal, discrimination reversal, context change, instructions, matching-tosample, humans ********** The study of the retrieval mechanisms of performance after retroactive interference treatments has received increasing attention during the last decade, both in human and nonhuman animals (see for instance Bouton, 1993; Bouton, Nelson, & Rosas, 1999; Romero, Vila, & Rosas, 2003; Rosas, Vila, Lugo, & Lopez, 2001; Vila, Romero, & Rosas, 2002). These studies show that the interference treatment does not eliminate first-learned information. Performance according to originally learned information recovers when the test is conducted some days after training (i.e., Spontaneous recovery; e.g., Pavlov, 1927; Rosas et al., 2001; Vila & Rosas, 2001), or in a context different from the training context (i.e., Renewal; e.g., Bouton & Bolles, 1979; Romero et al., 2003; Rosas, et al., 2001). To explain these results, Bouton (1993) suggested that interference treatments made ambiguous the information presented to the subjects. According to this author, this ambiguity leads subjects to pay attention to the context where such information was learned, so that the interfering information becomes context dependent. Any change in the context where the interference takes place would then lead to forgetting of the interfering information, and subsequent retrieval of first-learned information. According to this author, subjects similarly code physical and temporal contexts, explaining the effects of the passage of time and the physical context change with the same mechanism (see also Bouton et al., 1999; Rosas & Bouton, 1997, 1998; Rosas et al., 2001). Stimuli used as contexts in the human literature experiments were physical changes in the configuration of the computer screen where participants received the task, or in the rooms and the computers where participants were trained. In short, those context changes can be identified as physical contexts. The aim of the experiments reported here was to explore the possibility of using instructional contexts to control retrieval of performance in a discrimination reversal situation. The role of instructions in human performance has received large attention within the behavioral literature. Participants seem to behave according to the instructions given to them, controlling participants' performance under different reinforcement schedules, and extinction, leading to insensitivity to the actual response-outcome contingencies (e.g., Baron & Galizio, 1983; Baron, Kaufman, & Stauber, 1969; Bijou, 1958; Colgan, 1970; Hayes, Brownstein, Haas, & Greenway, 1986; Weiner, 1970). …