The Gleason score (GS) of prostate cancer is calculated by adding primary and secondary Gleason grades with patterns occupying less than 5% of the tumour often not included despite their probable prognostic significance. A modified Gleason score (mGS) comprising primary and tertiary patterns of higher grade has been proposed, but its interobserver variability has yet to be elucidated. Slides from 69 consecutive prostatectomy specimens were circulated among four genitourinary pathologists. GS and mGS were assessed, and results were compared in pairs. Mean weighted kappa for GS and mGS were 0.56 (range 0.52-0.66) and 0.58 (range 0.49-0.74), respectively. The difference between GS and mGS was 0, 1 and 2 score units in 66%, 26% and 8%, respectively, mean 0.41 score units (range 0.24-0.51). The increment was greater for transition-zone tumours than for peripheral-zone tumours (0.63 and 0.35 score units, respectively, P=0.002). An odd mGS (5, 7 or 9) was more often given than an odd GS (77% and 62%, respectively, P<0.001). Disagreement between observers greater than 1 score unit was more common with mGS than GS (18% and 4%, respectively, P<0.001). In conclusion, overall mean weighted kappa for interobserver reproducibility of mGS is at least as high as that of GS. However, there is a clustering of mGS in odd scores, and severe disagreement is more commonly observed than with GS. Training of mGS assessment would possibly improve agreement. Tertiary Gleason patterns need to be better defined.