Inter-observer variation in cytohistological diagnosis was assessed for 1506 cervical smears and 883 histological slides from four case-control studies on cervical neoplasia. The kappa statistic among a panel of three cytopathologists was highest for diagnosis of invasive cancer (0.70 for cytology and 0.74 for histology), followed by normal/inflammatory in cytology (0.68) and CIN III in histology (0.58). There was also nearly perfect agreement between the final panel diagnoses and the original diagnoses made by local cytopathologists, except for those of CIN III. Inter-observer variation in diagnosis for CIN III was inversely associated with age, number of children (in histology) and sexual activity (in cytology). However, the odds ratios for CIN III calculated by each cytopathologist's diagnosis were not different from each other for any etiologic factor. These results indicate that the diagnoses of invasive cancer and of normal/inflammatory changes are highly reproducible and that the inter-observer variation does not have much impact on the etiologic risk estimates.