BACKGROUND: Identifying which categories in the World Health Organization classification of endometrial hyperplasia contribute to suboptimal reproducibility is clinically important. METHODS: A 2-member panel reviewed 209 endometrial biopsy/curettage specimens originally diagnosed as incident endometrial hyperplasia as part of a progression study. Original diagnoses included the following: disordered proliferative endometrium, simple hyperplasia, complex hyperplasia, and atypical hyperplasia; panel diagnoses also included negative and carcinoma. We assessed percentage agreement and kappa statistics+/-standard errors (K+/-SE). RESULTS: Original and panel diagnoses (combining negative with disordered proliferative endometrium; atypical hyperplasia with carcinoma) agreed for 34.9% of biopsies (K-unweighted+/-SE=0.18+/-0.03; K-weighted+/-SE=0.27+/-0.04). Panelists' diagnoses agreed (using 6 categories) for 51.7% of biopsies, corresponding to K-unweighted+/-SE=0.37+/-0.03, improving with weighting to K-weighted+/-SE=0.63+/-0.05. Reproducibility based on a 2-tier classification ([negative, disordered proliferative endometrium, simple hyperplasia, or complex hyperplasia] versus [atypical hyperplasia or carcinoma]) increased agreement between original and panel diagnoses to 82.8%, K-unweighted+/-SE=0.37+/-0.06, and between panelists to 87.0%, K-unweighted+/-SE=0.63+/-0.07. Agreement between panelists at a cutpoint of complex hyperplasia and more severe versus simple hyperplasia or less severe was 88.0%, K-unweighted+/-SE=0.72+/-0.07. CONCLUSIONS: Developing and prospectively testing a binary system of classifying endometrial hyperplasia on endometrial biopsy may aid efforts to improve interobserver reproducibility.