Simulations were performed to verify the accuracy with which the Mantel-Haenszel (MH) and Rasch PROX procedures recover simulated item bias. Several standard error estimators for the MH procedure were evaluated. Item bias is recovered satisfactorily by both techniques under all simulated conditions. The proposed MH standard error estimators have different properties and are to be interpreted with caution. When using the MH procedure, the analyst is advised to calculate the size of the item bias and error variance using several sub-setting criteria and more than one standard error estimator. The PROX procedure provides a computationally independent and statistically conservative check on the estimates provided by the MH procedure. (Six tables show variations of item bias estimates across new score sub-sets.) (Author)