An increasing number of genetically defined types of spinocerebellar ataxia (SCA) have been reported in the past decade. Phenotype--genotype correlation studies have suggested a broad overlap between SCA types. The aim of the present study was to identify patterns of clinical features that were likely to distinguish between SCA types and to test the specificity and sensitivity of these signs and symptoms using a Bayesian classifier. In total, 127 patients from 50 families with SCA types 1 to 8 were examined using a worksheet with a panel of 33 symptoms and signs. By computing the probabilities of each trait for each SCA type, we rated the predictive value of each feature for each form of ataxia and then combined the probabilities for the entire panel of traits to construct a Bayesian classifier. Results of this analysis were summarized in a simpler, more operator-based algorithm. Patients with SCA5, SCA6, and SCA8 demonstrated a predominant cerebellar syndrome, whereas patients with SCA1, SCA2, SCA3, SCA4, and SCA7 frequently had clinical features indicating an extracerebellar involvement. The Bayesian classifier predicted the SCA type in 78% of patients with sensitivities between 60 and 100% and specificities between 94 and 98.2%. The highest sensitivity to correctly predict the true SCA type was found for SCA5, SCA7, and SCA8. Sensitivities and specificities found in the present study validate the use of algorithms to help to prioritize specific SCA gene testing, which will help to reduce costs for gene testing.