This article follows a three-stage data envelopment analysis (DEA) approach proposed by Fried et al. (2002) to decompose mutual fund underperformance, in order to obtain pure managerial performance. In the first stage, DEA is used to compute each fund's performance. In the second stage, a stochastic frontier regression decomposes fund underperformance into characteristics (including fund and management attributes), managerial inefficiency, and statistical noise. In the third stage, DEA with slack-adjusted data is used to find out the pure performance. It is found that a fund's performance significantly increases with its size, previous performance, manager's tenure and education, while it decreases with the age of the fund and number of managed funds.