In an effort to identify reliable criteria for detecting seminal vesicle invasion (SVI) with transrectal ultrasonography (TRUS) in patients with clinically localised prostate cancer, we reviewed the pre-operative sonograms in 230 patients who underwent radical retropubic prostatectomy; 49 patients (21%) had pathologically confirmed SVI. Conventional sonographic criteria for SVI (asymmetry, distension, atrophy, abnormal echogenicity and irregularity in outline) were present in 58 patients, but only 16 (28%) had pathologically confirmed SVI. On the basis of the results of a preliminary comparison of radical prostatectomy specimens and TRUS, we had revised our criteria for the recognition of SVI: (1) a hypoechoic lesion at the base of the prostate (within 10 mm of the seminal vesicle); (2) an "adhesion sign" resulting from the loss of the echo reflections from the normal fat plane between the prostate and the seminal vesicle; (3) "posterior convexity" of the seminal vesicles. When we reviewed the 230 sonograms retrospectively, we found a hypoechoic tumour at the base in 70 patients, of whom 37 had SVI (positive predictive value (PPV) 53%). An adhesion sign was found in 16 patients, 12 of whom had SVI (PPV 75%). Posterior convexity was present in 4 patients, all of whom had SVI. If any one of our sonographic signs was present, the overall accuracy (83%), sensitivity (90%) and positive predictive value (51%) were significantly better than with any one of the conventional criteria. Patients with SVI were also more likely to have a high serum prostate specific antigen (PSA) level. The PPV for SVI of a PSA level > or = 10 ng/ml was 38%. If the PSA was > 10 ng/ml and TRUS was positive (> or = 1 of our sonographic criteria), 16 (62%) of 26 patients had SVI. If the PSA was < 10 ng/ml and TRUS was negative, only 3 (3%) of 86 patients had SVI. It was concluded that the conventional criteria for detecting SVI on ultrasonography are not accurate in patients with early stage prostate cancer. There are, however, reliable criteria that predict SVI with reasonable accuracy and these criteria, combined with the serum PSA levels, can stratify patients into those with a low risk and those with a high risk of SVI.