Recovery is dynamic during acute stroke, but whether new motor skills can be acquired with the paretic upper limb (UL) during this recovery period is unknown. Clarifying this unknown is important, because neurorehabilitation largely relies on motor learning. The aim was to investigate whether, during acute stroke, patients achieved motor skill learning and retention with the paretic UL. Over 3 consecutive days (D1–D3), 14 patients practiced with their paretic UL the CIRCUIT, a motor skill learning task with a speed/accuracy trade-off (SAT). A Learning Index (LI) was used to quantify normalised SAT changes in comparison with baseline. Spontaneous motor recovery was quantified by another task without SAT constraint (EASY), by grip force (GF), and the Box and Blocks test (BBT). In patients, CIRCUIT LI improved 98% ± 66.2 (mean ± SD). This improvement was similar to that of young healthy individuals (n = 30) who trained with a slightly different protocol for 3 consecutive days (83.8% ± 58.8%). Generalisation of SAT gains to an untrained circuit was observed in both groups. From D1 to D3, stroke patients improved their performance on EASY, while changes in GF and BBT were heterogeneous. During acute stroke, patients retained SAT gains for a motor skill learned with the paretic UL in a manner similar to that of healthy individuals. These results demonstrate acute stroke patients achieved motor skill learning and retention that exceeded paretic UL improvements explained by spontaneous recovery.