The persistence of mutual fund alphas is often viewed as evidence that some funds possess skill and that this skill persists. This interpretation is unwarranted when security factor lines are too flat. A sort on estimated alphas is a sort against betas: high-alpha funds are predominantly low-beta funds and vice versa. Thus, a strategy of investing in high-alpha funds benefits not from skill, but from a betting-against-multiple-betas effect. In-sample tests intended to separate luck from skill are similarly biased. We demonstrate this bias through simulations and show that applying a correction factor eliminates much of the apparent skill in mutual fund data.