The agreement between Mantel-Haenszel and Rasch procedures for identifying differential item functioning (DIF) on teacher certification tests was studied. Two specific research questions were addressed: (1) whether the Mantel-Haenszel and Rasch procedures identify the same items as functioning differently; and (2) how consistently each method identifies items with DIF over administrations. The sample included all black and white examinees who took one of the Georgia Teacher Certification Tests during the December (1987), March (1988), or June (1988) administrations. Item data from these three administrations within the content fields of early childhood (n=1,344; n=1,291; and n=1,023, respectively), middle childhood (n=1,009; n=845; and n=785, respectively), and administration and supervision (n=220; n=216; and n=252, respectively) were used in the analyses, and the differential performance of black and white examinees on these items was examined. The agreement between the two procedures was fairly high within the three administrations, but it dropped significantly when common items were examined across the administrations. The reliability of each procedure was also examined. The data suggest that the Rasch procedure is more consistent in identifying items with DIF than is the Mantel-Haenszel procedure. Further, the data suggest that quantitative indices of DIF are preferable to categorical indices for both procedures. Promising areas for future research on DIF are discussed, and the implications of the findings for theory and practice within the context of teacher certification tests are presented. Six tables present study data. (SLD)