In 3 experiments, participants, on signal, moved a cursor from a central position to 1 of 8 numerically labeled locations on the circumference of a clock face. Movements were controlled by a mouse in 1 of 4 conditions: vertical reversal, horizontal reversal, combined reversals, or normal (i.e., no reversals). Participants were trained in 1, 2, or 3 of these conditions and were tested 1 week later with either the same or a different condition. There were improvements across training and perfect retention across the delay. There was little or no transfer, however, even when training involved combined reversals or multiple conditions. These results illustrate severe specificity of training and are interpreted in terms of acquired inhibition of normal responses.