We develop a methodology for estimating bias-corrected premium estimates from cross-sectional regressions of individual stock returns on time-varying conditional betas. For a comprehensive sample of stocks over the post-war period from 1946 through 2011, we find fairly consistent evidence of a positive risk premium on the size factor, but limited evidence for the book-to-market and momentum factors (none for the market factor). Firm characteristics explain a much larger proportion of variation in estimated expected returns than factor loadings when return premia are taken to be constant. However, we find evidence of predictability in the premia for characteristics as well as loadings. Taking this into account, the gap between characteristics and loadings narrows (56% versus 44%) for the three-factor model and loadings take the lead (56% to 39%) with the addition of the momentum factor.