Our objective was to systematically review the existing literature regarding the use of cytokeratin (CK) stain in differentiating Barrett’s esophagus (BE) from tissues of the gastric cardia, corpus, or antrum, with or without intestinal metaplasia (IM). Pubmed was searched for full publications in English (1983–2005) addressing the use of CK for differentiation of BE from contiguous tissues. Information was collected on the study sample, blinding, the methods used for CK staining, and for defining and applying the gold standard tests. Test characteristics were obtained or calculated. Sixteen studies (containing 46 comparisons) met the inclusion and exclusion criteria. Immunostaining for CK 7 and 20 was generally highly specific in distinguishing long-segment BE from antrum IM, fundus IM, or noncardiac gastric IM; 27 comparisons showed statistically significant differences. However, only 8 of 15 comparisons (6 of 12 studies) reported significant differences in CK staining patterns between BE and gastric cardia IM with a high sensitivity (89%–100%) and specificity (83%–100%) for long-segment BE and lower estimates for short-segment BE, while the other seven comparisons showed no significant differences and a very low sensitivity. Examination by a blinded pathologist was reported in five of six positive studies and in only one of six of the negative studies. In addition, variation in the patient populations, use of surgical resection versus endoscopic biopsies, and biopsy sampling technique in endoscopic studies may have accounted for these differences. Finally, two studies did not find significant differences in CK staining patterns between BE and normal cardiac mucosa. In conclusions, CK immunostaining has not performed well in differentiating BE, especially short-segment BE, from cardia IM. There seems to be a spectrum bias where the accuracy varies with different tested populations. CK immunostaining distinguished well between BE and IM in noncardiac segments of the stomach; however, these comparisons are not clinically relevant.