We present a new approach to selecting actively managed mutual funds that uses both portfolio holdings and fund return information to eliminate funds with predicted inferior performance through a sequence of pairwise fund comparisons. Our methodology determines both the number of skilled funds and their identities, and locates funds with substantially higher risk-adjusted returns than those identified by conventional alpha-ranking methods. We find strong evidence of time-series variation in both the number of funds identified as superior using our approach, as well as in their performance across different economic states.