WWe re-visit a puzzling result that in U.S. post-WW II data the dividend price ratio can predict aggregate returns but not dividend growth. We find that predictive regressions are sensitive to the method used to aggregate firm-level data. Using value weighted firm-level data we find strong evidence for dividend growth predictability in the post-WW II period. We explore the reasons behind the differences in predictability due to different weighting methods. We find that these differences in predictability are related to the fact, in the data, that it is not always the largest firms that pay the largest dollar dividends or earnings.