In an investigation of item bias, the stability of Mantel-Haenszel (MH) estimates across different samples of test takers and different sample sizes and the robustness of the MH procedure with respect to item context effects were investigated. Data from the Second International Mathematics Study (1985) were analyzed. The data consisted of responses to 40 core items on the mathematics tests and one of four 35-item rotated mathematics tests for a core sample of 670 black and 5,015 white eighth graders in the United States. Most analyses were performed with a sample of just over 100 blacks and slightly over 1,000 whites. Correlations between different samples of test takers were low, suggesting that relatively larger sample sizes were necessary for stable estimates. Correlational analyses also suggest that the MH procedure was robust to item context effects. Four of the 40 core items were identified as functioning differently for black and white test takers. The MH procedure is useful in detecting differential item functioning; however, further investigation is needed to determine the sample size necessary for stable estimates. Seven tables present study results. A 21-item list of references is included. (SLD)