Researchers are increasingly using data from the Nasdaqmarket to examine pricing behavior, market design,and other microstructure phenomena. The validity ofany study that classifies trades as buys or sellsdepends on the accuracy of the classificationmethod. Using a Nasdaq proprietary data set thatidentifies trade direction, we examine the validityof several trade classification algorithms. We findthat the quote rule, the tick rule, and the Lee andReady (1991) rule correctly classify 76.4%, 77.66%,and 81.05% of the trades, respectively. However, allclassification rules have only a very limitedsuccess in classifying trades executed inside thequotes, introducing a bias in the accuracy ofclassifying large trades, trades during high volumeperiods, and ECN trades. We also find that extantalgorithms do a mediocre job when used forcalculating effective spreads. For Nasdaq trades, wepropose a new and simple classification algorithmthat improves over extant algorithms.