This study investigated practice effects on reaction time (RT) in patients with unilateral brain damage and age-matched controls. Subjects performed a Sternberg-type stimulus recognition task. Both RT and RT variability decreased in all subjects from a short initial session to a second session, 1 to 3 days later. Within the second session, however, RT remained constant in the controls, whereas it was further reduced in the patients. These findings argue against the assumption that brain-damaged patients generally show smaller practice effects on RT than controls, and instead suggest that it may be essential to provide patients with sufficient practice opportunities. To locate the processing stage where practice is effective, analyses were performed also on the intercept and slope parameters of the RT function. It is argued that the differential practice effects were due to the speeding of perceptual and/or response-related stages, rather than gains in memory search speed.