The paper uses meta-analysis to investigate the evolution of returns to education in Russia during the economic transition. We present the evidence of strong increase in returns to education in the 1990s. The returns to education peaked at 8% per additional year of schooling in the early 2000s. Since the mid-2000s the positive trend has been reversed. We find that the estimated returns are sensitive to sample design, spe­cification and estimation methods. On average the estimated returns are higher if one uses imputed (instead of actual) years of education and estimates earnings equation with the OLS. Among covariates, the form of ownership and settlement type (urban/ rural) has the largest effect on the estimated returns to education. Regional variables have significant impact on the estimated returns only being included into the earnings equation at the oblast (province) level.