Glomerular sclerosis, mesangial hypercellularity, extracapillary lesions, interstitial fibrosis, and vascular sclerosis have been reported to be the significant pathologic prognosticators in IgA nephropathy (IgAN). We developed our own scoring for the following main glomerular changes in 248 patients with IgAN: 1) glomerular hypercellularity (mesangial and endocapillary), 2) segmental lesions such as tuft adhesion, crescent and segmental sclerosis, 3) global glomerular sclerosis. Indices of each lesion were semiquantitatively determined. The sum of these three indices was defined to be a glomerular score. We found that a glomerular score significantly related to the outcome of patients with IgAN in univariate life table analysis. We also semiquantitatively determined total score including tubulo-interstitial and vascular lesions as well as glomerular score and compared the predictive power as a prognosticator between glomerular score and total score. Using Cox's proportional Hazard model and log-likelihood ratio test, we confirmed that predictive power of glomerular score was better than that of total score. Furthermore, we assessed the reproducibility of glomerular score using Kappa statistics. Three pathologists read 100 biopsies which were randomly selected from the materials and all pathologists read them twice. A value of Kappa between the first and second observation of pathologist A, B and C was 0.68, 0.71 and 0.60, respectively. Values of Kappa between Pathologist A and B were ranging from 0.45 to 0.47, those between Pathologist A and C from 0.30 to 0.36, and finally those between Pathologist B and C were ranging from 0.12 to 0.23. Therefore, intra-observer reproducibility was nearly excellent. And inter-observer reproducibility between Pathologist A and B was satisfactory. However, inter-observer reproducibility between Pathologist A and C and between B and C was not satisfactory. We feel our scoring system is very convenient and easy to be understood as a prognosticator in patients with IgAN. It, however, should be used by one pathologist because of excellent intra-observer reproducibility and rather unsatisfactory inter-observer reproducibility.