We assessed the validity and interrater reliability of neurologists who, using four different sets of previously published criteria for the clinical diagnosis of progressive supranuclear palsy (PSP), also called Steele-Richardson-Olszewski syndrome, rated 105 autopsy-proven cases of PSP (n = 24), Lewy body disease (n = 29), corticobasal ganglionic degeneration (n = 10), postencephalitic parkinsonism (n = 7), multiple system atrophy (n = 16), Pick's disease (n = 7), and other parkinsonian or dementia disorders (n = 12). Cases were presented in random order to six neurologists. Information from each patient's first and last visits to the medical center supplying the case was presented sequentially to the rater, and the rater's diagnosis was compared with the neuropathologic diagnosis of each case. Interrater agreement for the diagnosis of PSP varied from substantial to near perfect, but none of the criteria had both high sensitivity and high predictive value. Because of these limitations, we used a logistic regression analysis to identify the variables from the data set that would best predict the diagnosis. This analysis identified vertical supranuclear palsy with downward gaze abnormalities and postural instability with unexplained falls as the best features for predicting the diagnosis. From the results of the regression analysis and the addition of exclusionary features, we propose optimal criteria for the clinical diagnosis of PSP.