Although the characteristic of invasive pattern which contributes to Jass's classification is a sensitive prognostic marker in rectal cancer, reproducibility of its assessment has been shown to be problematic. As another histological parameter of invasive margin, we examined the prognostic significance of tumour 'budding' and attempted to establish its appropriate criteria. A total of 638 rectal cancer specimens was examined. We defined tumour 'budding' as an isolated single cancer cell or a cluster composed of fewer than five cancer cells. We divided these into two groups by their intensity, i.e. the number of 'budding' foci within a microscopic field of x 250. Rectal cancer with high-grade 'budding' (>or= 10 foci in a field) was observed in 30.1% of patients, and was associated with lower 5-year survival rates (40.7%) than patients with low-grade 'budding' (84.0%) (P < 0.0001). Based on multivariate analysis, tumour 'budding' was selected as the significant independent variable, together with the number of nodes involved, extramural spread, lymphocytic infiltration, apical nodal involvement and tumour differentiation. Kappa coefficient of two-graded tumour 'budding' in the intraobserver study was 0.84. Because of its value as a prognostic indicator and its reproducibility, tumour 'budding' would be a good index to estimate the aggressiveness of rectal cancer.