Four item bias methods were studied. The methods compared include the Mantel-Haenszel statistic, the plot method, the route mean squared difference method, and the total area method; the latter two methods are based on item response theory. The test consisted of item responses of 451 male and 486 female ninth graders to 75 test items on the 1985 Cleveland Reading Competency Test. Focus was on sex bias. Simulated data were used to set cut-off scores for interpreting the item bias statistics. Each method led to the identification of nearly the same set of potentially biased items. Methodological problems included imprecision in establishing cut-off points, Type I errors, and poor item parameter estimates. Results highlight the importance of the choice of interval on the ability scale over which item bias is measured. It appears that the Mantel-Haenszel statistic provides a quick, cheap alternative to the more laborious and expensive item response theory methods. Seven graphs and four tables are included. (TJH)