We performed a pilot randomized, controlled trial of intensive, computer-based cognitive training in 47 subjects with mild cognitive impairment. The intervention group performed exercises specifically designed to improve auditory processing speed and accuracy for 100 min/d, 5 d/wk for 6 weeks; the control group performed more passive computer activities (reading, listening, visuospatial game) for similar amounts of time. Subjects had a mean age of 74 years and 60% were men; 77% successfully completed training. On our primary outcome, Repeatable Battery for Assessment of Neuropsychological Status total scores improved 0.36 standard deviations (SD) in the intervention group (P=0.097) compared with 0.03 SD in the control group (P=0.88) for a nonsignificant difference between the groups of 0.33 SD (P=0.26). On 12 secondary outcome measures, most differences between the groups were not statistically significant. However, we observed a pattern in which effect sizes for verbal learning and memory measures tended to favor the intervention group whereas effect sizes for language and visuospatial function measures tended to favor the control group, which raises the possibility that these training programs may have domain-specific effects. We conclude that intensive, computer-based mental activity is feasible in subjects with mild cognitive impairment and that larger trials are warranted.