We show that the dividend yield and earnings yield jointly are strong predictors of dividend growth. We motivate the joint specification with a theoretical model and show how omitting the earnings yield biases the dividend yield coefficient towards zero, explaining why the dividend yield by itself is a poor predictor of dividend growth. Our empirical results are robust in pre- and post-war U.S. data, in recessions and expansions, in international data, and when controlling for additional predictors.