OBJECTIVE To determine if adults in the subacute phase post moderate or mild stroke could learn a motor task under implicit practice conditions. METHODS Non-randomized, between-group design. METHODS Subjects' homes. METHODS Sample of convenience of 15 adults with moderate stroke, 22 adults with mild stroke, and 32 age-matched controls. Stroke severity was determined using the Orpington Prognostic Scale. METHODS Practice of movements to target switches in two conditions, a repeated series and a random series. Practice was organized into a single session of six blocks of 80 trials such that blocks 1 and 2 were the random series, blocks 3 and 4 the repeated series, block 5 the random series, and block 6 the repeated series. Explicit knowledge of the two conditions was not provided. METHODS The mean response time and the coefficient of variation were calculated for each block for each group. RESULTS Regardless of group, participants had a slower response time when practice was unexpectedly switched from the repeated to random condition. The response time and coefficient of variation for those with moderate stroke were persistently greater than those with mild stroke and controls, who were not different from each other. CONCLUSIONS Implicit motor learning is preserved in adults with moderate stroke and may be a viable strategy for teaching motor skills in rehabilitation. The clinician should be aware that subjects with moderate stroke, even when performing with the less affected upper extremity, have significantly slower and more variable movements than those with mild stroke and controls.