Discrimination learning and reversal problems were given to 128 children aged two and three years in which there were either two, three, six, or nine choice-alternatives per problem. The position of the correct object (either a toy car on the original problem and a ball on the reversal problem or vice versa) was varied randomly among all the possible positions, all other positions being occupied by duplicates of the incorrect object. It was found that two-choice reversal problems were much more difficult than all others. There was also a direct and monotonic relationship between ease of solution and number of choices present during reversal learning. Additional analysis inlicated that position response biases were probably responsible for this effect.