Background: Hand motor deficits contribute to multiple sclerosis (MS)-related disability. Action observation training (AOT) is promising to improve upper limb function in neurologic patients. Objectives: In this preliminary study, we investigated AOT effects on dominant-hand motor performance in MS patients with upper limb motor impairment and performed an explorative analysis of their anatomical and functional magnetic resonance imaging (MRI) substrates. Methods: In total, 46 healthy controls (HC) and 41 MS patients with dominant-hand motor impairment were randomized to AOT (HC-AOT = 23; MS-AOT = 20; watching daily-life action videos and execution) or control-training (HC-Control = 23; MS-Control = 21; watching landscapes videos and execution). Behavioral, structural, and functional (at rest and during object manipulation) MRI scans were acquired before and after a 2-week training. Results: After training, MS groups improved in right upper limb functions, mainly in AOT group ( p from 0.02 to 0.0001). All groups showed regional increased and decreased gray matter volume, with specific AOT effects in fronto-temporal areas in MS-AOT ( p < 0.001), without white matter (WM) integrity modifications. Increased and reduced recruitments of the action observation matching system and its connections in MS-AOT were found ( p < 0.001). Motor improvements were correlated with volumetric and functional MRI modifications ( r from −0.78 to 0.77, p < 0.001). Conclusion: The 10-day AOT promotes clinical improvements in MS patients through structural and functional modifications of the action observation matching system.