Recent studies based on multivariate analysis have identified cytologic features that may be of value in the diagnosis of malignancy in bile duct brushings. We sought to assess the reproducibility and accuracy of these criteria. Three different observers used 4 sets of criteria that included 9 cytologic features to review 165 bile duct brushing specimens with available follow-up data. An overall assessment of malignancy and evaluation for the presence of chromatin clumping had excellent reproducibility (3-way kappa values of 0.708 and 0.629, respectively). Evaluation for the presence of enlarged nuclei, increased nuclear:cytoplasmic ratio, nuclear molding, and loss of honeycombing showed moderate reproducibility. An overall assessment of malignancy was a better predictor of the presence of malignancy than any other criteria, with a sensitivity of 36.2% and a specificity of 95%. Retrospective analysis demonstrated that the criteria of chromatin clumping, increased nuclear:cytoplasmic ratio, and either nuclear molding or loss of honeycombing had similar sensitivity and specificity as the overall assessment of malignancy (sensitivity = 35.2%, specificity = 95%). An overall assessment of malignancy or the criteria of chromatin clumping, increased nuclear:cytoplasmic ratio, and either nuclear molding or loss of honeycombing are reproducible cytologic criteria that accurately predict malignancy in bile duct brushings.