This paper shows that short horizon stock returns can be predicted to a much greater degree by past price movements than would be anticipated given their low autocorrelation. This raises doubts over the reliability of the autocorrelation statistic as a measure of stock market predictability.