The hypothesis that task attentional requirements vary inversely with increases in the efficiency of task performance was evaluated. Three experimental groups (N = 8, N = 6, N = 3) were required to learn a two-hand coordination task (THC). To assess changing attentional demands at different levels of skilled THC performance, subjects simultaneously performed a random number generation task (RNG) on two or four trials of the THC task; these trials occurred during initial acquisition, during semiskilled performance, at performance asymptote, or after overlearning. The RNG index showed marked deterioration during initial acquisition (p <.005), remained below baseline but improved during semiskilled performance (p <.05), and partially and completely recovered to original baseline levels during the mastery and overlearning trials. The RNG procedure may provide a brief, sensitive, and consistent measure of attention deployment during the performance of tasks and in the learning and acquisition of complex skills.