We measure the postsplit performance of 12,747 stock splits from 1927 to 1996 using two methods to measure abnormal returns: size and book-to-market reference portfolios with bootstrapping, and calendar-time abnormal returns combined with factor models. Between 1927 and 1996, neither method applied to splits 25 percent or larger finds performance significantly different from zero. Over selected subperiods, subsamples of 2-1 splits restricted by book-to-market availability requirements display positive abnormal returns using some methods. However, these samples show small or negligible abnormal returns using the calendar-time method. Overall, the stock split evidence against market efficiency is neither pervasive nor compelling.