Rectal mucosal proliferation has been shown to be increased in patients with neoplastic lesions of the large bowel and may serve as a marker of risk for colorectal malignancy. We conducted analyses to determine reliability and components of variability that might suggest optimal analysis strategies for studies of proliferation. Endoscopic pinch biopsies were obtained from 17 adult patients, labeled using proliferating cell nuclear antigen, scored using strict rules, and then rescored. Labeling index, defined as the proportion of labeled cells in a crypt, was calculated for each crypt, biopsy, subject, and group. There was excellent reproducibility. The technician was able to select previously scored crypts 95% of the time. The overall labeling index was identical on repeat. There was considerable variability in labeling index among crypts from a single biopsy and between biopsies of a single subject. Variance component estimates suggested that 20% of the variability of labeling index was due to subject, 30% due to the biopsy within a subject, and 50% due to crypts within a biopsy. There were substantial gains in statistical power by scoring two biopsies rather than one. There was less gain from further increases in biopsy number. There was little statistical advantage for counting more than 8 crypts/biopsy. Demonstrating a decrease of 25% in the mean labeling index with 90% power could require more than 100 subjects/group. We conclude that proliferating cell nuclear antigen is an extremely reproducible method to determine proliferation index. There is considerable variability among subjects, biopsies, and crypts.(ABSTRACT TRUNCATED AT 250 WORDS)