I find that limit order submission is not random. The probability of observing a limit sell (buy) order is highest after another limit sell (buy) order. This first-order positive serial correlation in the order flow cannot be explained by theories offered in the current literature. I find that the positive serial correlation is due largely to order splitting and the undercutting behavior among limit order traders. Controlling for these two effects explains up to 72 percent of the positive serial correlation. I also show that the failure to account for these two effects may cloud the results of some empirical market microstructure studies.