Substance dependence is thought to be mediated by abnormalities in cognitive abilities, but how this impacts decision-making remains unclear. This study aimed to test whether people who are opiate dependent differed from never-dependent controls in learning from reward and punishment or in the generalization of learning to novel conditions. Participants with opiate dependency consisted of 21 people who were outpatients in a methadone maintenance program; the control group consisted of 21 healthy participants with no histories of substance abuse. Subjects completed a computer-based task that involved two phases: the training phase involved participants being presented with compound stimulus (a shape and color) in each trial, with the goal of learning which compounds to ‘pick’ for rewards or ‘skip’ to avoid punishment. The test phase involved a transfer test, where stimuli from the first phase were combined together to form novel compounds without feedback. The control group demonstrated fewer errors compared to opiate-dependent individuals during the training phase. In the test phase, controls used prior knowledge of both shapes and colors in responding; however, opiate-dependent individuals used shapes but did not use their knowledge of color to modulate responding. When performance during training was equated in the groups using a learning threshold, this difference between groups on the generalization test remained. A deficit in learning generalization might be indicative of group differences in learning strategies in operation during training; however, future work is necessary to uncover the specific neural substrates in action during transfer tasks, and to determine the effects of acute methadone dosage on decision-making.