Ataxia and impaired motor learning are both fundamental features in diseases affecting the cerebellum. However, it remains unclarified whether motor learning is impaired only when ataxia clearly manifests, nor it is known whether the progression of ataxia, the speed of which often varies among patients with the same disease, can be monitored by examining motor learning. We evaluated motor learning and ataxia at intervals of several months in 40 patients with degenerative conditions [i.e., multiple system atrophy (MSA), Machado–Joseph disease (MJD)/spinocerebellar ataxia type 3 (SCA3), SCA6, and SCA31]. Motor learning was quantified as the adaptability index (AI) in the prism adaptation task and ataxia was scored using the Scale for the Assessment and Rating of Ataxia (SARA). We found that AI decreased most markedly in both MSA-C and MSA-P, moderately in MJD, and mildly in SCA6 and SCA31. Overall, the AI decrease occurred more rapidly than the SARA score increase. Interestingly, AIs remained normal in purely parkinsonian MSA-P patients (n = 4), but they dropped into the ataxia range when these patients started to show ataxia. The decrease in AI during follow-up (dAI/dt) was significant in patients with SARA scores < 10.5 compared with patients with SARA scores ≥ 10.5, indicating that AI is particularly useful for diagnosing the earlier phase of cerebellar degeneration. We conclude that AI is a useful marker for progressions of cerebellar diseases, and that evaluating the motor learning of patients can be particularly valuable for detecting cerebellar impairment, which is often masked by parkinsonisms and other signs.