The effect of extensive training on the contribution of response--outcome learning to instrumental performance in rats was re-examined in two experiments using a transfer test. In each experiment, two discriminative stimuli were established as signals for different response--outcome combinations (e.g. light: nose poke-pellets and noise: handle pull-sucrose). Then, two different responses (lever press and chain pull) were concurrently trained with those outcomes. In Experiment 1, these responses underwent extensive training, each with a different outcome (e.g. lever press-pellets and chain pull-sucrose). In Experiment 2, these responses were trained moderately with one outcome (e.g. lever press-pellets and chain pull-sucrose) and extensively with a different outcome (e.g. lever press-sucrose and chain pull-pellets). Finally, transfer tests were conducted in which the discriminative stimuli, noise and light, were tested periodically with the lever and chain. In Experiment 1, the stimuli consistently and preferentially elevated performance of the response trained with the same outcome relative to that trained with a different outcome. In Experiment 2, the stimuli elevated both responses nondifferentially. However, a drive manipulation (thirst) designed to increase the value of sucrose relative to pellets revealed a significant preference for the response that had been extensively trained with sucrose. Overall, the results of these experiments confirm previous findings that instrumental behaviors do not become increasingly independent of their consequent outcomes with extended training. However, the transfer results of Experiment 2 highlight a potential limitation of the transfer test for assessing variations in the strength of R--O associations.