BACKGROUND: It is currently unknown whether motor skill learning (MSkL) with the paretic upper limb is possible during the acute phase after stroke and whether lesion localization impacts MSkL. Here, we investigated MSkL in acute (1–7 days post) stroke patients compared with healthy individuals (HIs) and in relation to voxel-based lesion symptom mapping. METHODS: Twenty patients with acute stroke and 35 HIs were trained over 3 consecutive days on a neurorehabilitation robot measuring speed, accuracy, and movement smoothness variables. Patients used their paretic upper limb and HI used their nondominant upper limb on an MSkL task involving a speed/accuracy trade-off. Generalization was evaluated on day 3. All patients underwent a 3-dimensional magnetic resonance imaging used for VSLM. RESULTS: Most patients achieved MSkL demonstrated by day-to-day retention and generalization of the newly learned skill on day 3. When comparing raw speed/accuracy trade-off values, HI achieved larger MSkL than patients. However, relative speed/accuracy trade-off values showed no significant differences in MSkL between patients and HI on day 3. In patients, MSkL progression correlated with acute motor and cognitive impairments. The voxel-based lesion symptom mapping showed that acute vascular damage to the thalamus or the posterior limb of the internal capsule reduced MSkL. CONCLUSIONS: Despite worse motor performance for acute stroke patients compared with HI, most patients were able to achieve MSkL with their paretic upper limb. Damage to the thalamus and posterior limb of the internal capsule, however, reduced MSkL. These data show that MSkL could be implemented into neurorehabilitation during the acute phase of stroke, particularly for patients without lesions to the thalamus and posterior limb of the internal capsule.