Based on the assumption that both sides of a gastric biopsy sample have a similar appearance, the choice of which side of a specimen is to be sectioned is usually random. We tested the hypothesis that the diagnosis reached by examining the two sides of a gastric biopsy may differ. Eighty-one antral biopsy specimens from patients with neither focal lesions nor previous gastric surgery were evaluated. After preparing 8-10 sections from each of the two opposite faces of each biopsy, sections were scored for Helicobacter pylori, activity, atrophy, intestinal metaplasia, lymphoid follicles. The presence of other lesions was also noted. Intrabiopsy agreement (the consistency in the identification of histological lesions on the opposite sides of the same sample) was calculated by using kappa statistics. A kappa value lower than 0.5 was considered "poor"; between 0.51 and 1.00 "moderate to excellent." The intrabiopsy agreement kappa values were: activity = 0.83, atrophy = 0.54, intestinal metaplasia = 0.51 and lymphoid follicles = 0.19. A mean of 42 serial sections (ranging from a mean of 22 for lymphoid follicles to a mean of 81 for xanthogranulomata) was needed to achieve an excellent (ie, kappa > or = 0.75) intrabiopsy agreement between the features showed at the opposite sides of a biopsy specimen. Intrabiopsy variability may represent a hitherto unrecognized source of error, and it should be minimized or avoided. In a research context information about the number of sections examined from a biopsy would provide a crucial element necessary to evaluate the accuracy of the histological data.