After integration into the endoplasmic reticulum (ER) membrane, ER-resident membrane proteins must be segregated from proteins that are exported to post-ER compartments. Here we analyze how human Gaa1 and PIG-T, two of the five subunits of the ER-localized glycosylphosphatidylinositol transamidase complex, are retained in the ER. Neither protein contains a known ER localization signal. Gaa1 is a polytopic membrane glycoprotein with a cytoplasmic N terminus and a large luminal loop between its first two transmembrane spans; PIG-T is a type I membrane glycoprotein. To simplify our analyses, we studied Gaa1 and PIG-T constructs that could not interact with other subunits of the transamidase. We now show that Gaa1(282), a truncated protein consisting of the first TM domain and luminal loop of Gaa1, is correctly oriented, N-glycosylated, and ER-localized. Removal of a potential ER localization signal in the form of a triple arginine cluster near the N terminus of Gaa1 or Gaa1(282) had no effect on ER localization. Fusion proteins consisting of different elements of Gaa1(282) appended to alpha2,6-sialyltransferase or transferrin receptor could exit the ER, indicating that Gaa1(282), and by implication Gaa1, does not contain any dominant ER-sorting determinants. The data suggest that Gaa1 is passively retained in the ER by a signalless mechanism. In contrast, similar analyses of PIG-T revealed that it is ER-localized because of information in its transmembrane span; fusion of the PIG-T transmembrane span to Tac antigen, a plasma membrane-localized protein, caused the fusion protein to remain in the ER. These data are discussed in the context of models that have been proposed to account for retention of ER membrane proteins.