The variation between three pathologists examining histological features seen in non-neoplastic colonic mucosa from 40 biopsies was analysed. Several procedures to express observer variation were used and compared, with emphasis on kappa statistics. Only five features, the presence of ulceration, villous surface, epithelioid granulomas, severe mucus depletion and crypt abscesses were sufficiently reproducible by the three pairs of pathologists. These findings suggest that other criteria used for the classification of inflammatory bowel disease are potentially unsatisfactory. When results from different studies on biopsies are being compared, influence of observer variation should be identified. Comparison of statistical techniques showed overall variation to be less useful than other statistical procedures. There was little difference between results from kappa statistics and other measures of agreement (overall agreement excluded).