Introduction. The minimal specimen size necessary for accurate interpretation of a renal biopsy has not been identified. We attempted such a determination by three different analyses of a collection of biopsies performed in renal transplants. Methods. First, we studied the influence of three lesions (glomerulosclerosis, arteriolar hyalinosis, interstitial fibrosis/tubular atrophy) in 199 baseline biopsies, obtained at time of transplantation, on transplant outcome. Secondly, we compared the results from the three lesions in baseline biopsy with those from 114 subsequent core biopsies in the same patients. Thirdly, we compared the two baseline biopsies obtained in 118 paired kidneys in cadaver transplantation where both kidneys were used. Results. For statistically significant prediction of outcome from glomerulosclerosis, we found that a specimen containing at least 25 glomeruli was needed in the baseline biopsy. Arteriolar hyalinosis predicted outcome independent of sample size, but became less important than percentage glomerulosclerosis in predicting outcome if only samples containing more than 25 glomeruli were considered. Interstitial fibrosis/tubular atrophy did not predict the outcome of a kidney, independent of sample size. When comparing baseline with subsequent core biopsies, or with paired baseline biopsies, at least 14 glomeruli were necessary to allow even moderate reproducibility of glomerulosclerosis (Cohen's kappa > 0.25) and to allow statistical significance (P < 0.05). The reproducibility of arteriolar hyalinosis was not dependent on sample size but was reproducible in 80% of paired baseline biopsies, and in 67% of the comparison of the baseline with core biopsy. Both precision and significance was lost if sample numbers were reduced by including only larger samples. There was no reproducibility in any study of interstitial fibrosis/tubular atrophy when comparing either baseline with subsequent biopsy, or paired baseline biopsies. Summary. Much larger biopsy samples are necessary than has generally been assumed in order for glomerulosclerosis rates to be reproducible or predictive of outcome. Arteriolar hyalinosis is prognostically important and shows good reproducibility independent of sample size. Interstitial fibrosis/tubular atrophy appear useless as predictors, being of no prognostic importance and lacking reproducibility. Our finding clarifies some of the discrepancies found by different investigators regarding the importance of renal biopsy in predicting prognosis. Preliminary, our data indicate that samples containing fewer than 25 glomeruli are unreliable in determining outcome based on glomerulosclerosis. The importance of our findings which are based only on chronic lesions, with respect to acute changes, is unknown.