We investigated the validity and reliability of diagnoses made by eight neuropathologists who used the preliminary NINDS neuropathologic diagnostic criteria for progressive supranuclear palsy (PSP) and related disorders. The specific disorders were typical, atypical, and combined PSP, postencephalitic parkinsonism, corticobasal ganglionic degeneration, and Pick's disease. These disorders were chosen because of the difficulties in their neuropathologic differentiation. We assessed validity by measuring sensitivity and positive predictive value. Reliability was evaluated by measuring pairwise and group agreement. From a total of 62 histologic cases, each neuropathologist independently classified 16 to 19 cases for the pairwise analysis and 5 to 6 cases for the group analysis. The neuropathologists were unaware of the study design, unfamiliar with the assigned cases, and initially had no clinical information about the cases. Our results showed that with routine sampling and staining methods, neuropathologic examination alone was not fully adequate for differentiating the disorders. The main difficulties were discriminating the subtypes of PSP and separating postencephalitic parkinsonism from PSP. Corticobasal ganglionic degeneration and Pick's disease were less difficult to distinguish from PSP. The addition of minimal clinical information contributed to the accuracy of the diagnosis. On the basis of results obtained, we propose clinicopathologic diagnostic criteria to improve on the NINDS criteria.