OBJECTIVE: To test the predictive ability of immunohistochemical estrogen receptor(ER), progesterone receptor (PR), COX-2, Mlh1, and Bcl-2 expressions for predicting the outcomes of regression and relapse in women with endometrial hyperplasia treated with the Levonorgestrel-releasing intrauterine system (LNG-IUS). METHODS: We recruited prospectively all women diagnosed with complex or atypical complex hyperplasia that underwent treatment with LNG-IUS from August 1998 until September 2008. Immunohistochemistry was performed with conventional methods and recorded using a semi-quantitative score (Q score) by two blinded assessors. Women were followed with endometrial biopsies to record regression and relapse. The biomarker predictive ability was analysed using the Cox proportional hazards model. RESULTS: The median follow-up was 72.1months (IQR 59.1-89.8). The Q score agreement between assessors was 82.6% (K statistic=0.801±0.036). The majority of study participants initially regressed to normal endometrium following LNG-IUS therapy (n=164 regressed; n=10 persisted). From the 164 women that regressed with LNG-IUS we were able to assess 152 women for relapse from which 18 relapsed. We found a weak association for persisted endometrial hyperplasia with ER and PR expressions with Q score on the 5th and 10th centiles. No associations were found for COX-2, Mlh1 and Bcl-2 protein expressions for regression and for any of the biomarkers for relapse. CONCLUSION: We found that poor expression of ER and PR is weakly associated with persisting endometrial hyperplasia and COX-2, Mlh1, and Bcl-2 expressions are not predictive. None of the biomarkers is predictive for relapse in women with endometrial hyperplasia treated with LNG-IUS.