OBJECTIVE To assess the diagnostic accuracy of transrectal ultrasonography (TRUS) in predicting the local extent of prostate cancer before radical prostatectomy. METHODS From April 1990 to April 1993, 59 consecutive patients (mean age 63 years, range 49-71) with clinically localized prostate cancer were examined using TRUS before radical prostatectomy. Primary tumours were clinically categorized according to the TNM classification and biopsies graded histologically. Tumour size was measured as length on a longitudinal and height and width on a transverse ultrasonogram. The results from TRUS were compared with those from the histopathological examination of whole-mount step sections of the surgical specimens. RESULTS In 50 patients, TRUS using standard criteria achieved sensitivity, specificity, positive- and negative predictive values (PPV, NPV) of 0.47, 0.63, 0.73 and 0.36, respectively, in predicting pathological stage. Dimensions were obtained from 45 patients with hypoechoic tumours; in a multivariate linear discriminant analysis the TRUS estimate of width was the best predictor of pathological stage. The sensitivity, specificity, PPV and NPV for tumour width from TRUS, using a threshold of 19 mm, were 0.42, 0.93, 0.93 and 0.43, respectively. CONCLUSIONS Tumour width measured by TRUS may improve the assessment of the local extent of prostate cancer.