We investigated in-sample and out-of-sample predictability of equal-weighted and capitalization-weighted quarterly excess returns for 55 industries over the 1973–95 period. The in-sample analysis supported predictability for about 80 percent of the cap-weighted industries and about 90 percent of the equal-weighted industries. The out-of-sample analysis provided strong evidence that the forecasting models for industry returns combined with mean–variance optimization criteria are useful for portfolio selection.