We studied the role of the association between glycosylphosphatidylinositol (GPI)-anchored proteins and glycosphingolipid (GSL) clusters in apical targeting using gD1-DAF, a GPI-anchored protein that is differentially sorted by three epithelial cell lines. Differently from MDCK cells, where both gD1-DAF and glucosylceramide (GlcCer) are sorted to the apical membrane, in MDCK Concanavalin A-resistant cells (MDCK-ConAr) gD1-DAF was mis-sorted to both surfaces, but GlcCer was still targeted to the apical surface. In both MDCK and MDCK-ConAr cells, gD1-DAF became associated with TX-100-insoluble GSL clusters during transport to the cell surface. In dramatic contrast with MDCK cells, the Fischer rat thyroid (FRT) cell line targeted both gD1-DAF and GlcCer basolaterally. The targeting differences for GSLs in FRT and MDCK cells cannot be accounted for by a differential ability to form clusters because, in spite of major differences in the GSL composition, both cell lines assembled GSLs into TX-100-insoluble complexes with identical isopycnic densities. Surprisingly, in FRT cells, gD1-DAF did not form clusters with GSLs and, therefore, remained completely soluble. This clustering defect in FRT cells correlated with the lack of expression of VIP21/caveolin, a protein localized to both the plasma membrane caveolae and the trans Golgi network. This suggests that VIP21/caveolin may have an important role in recruiting GPI-anchored proteins into GSL complexes necessary for their apical sorting. However, since MDCK-ConAr cells expressed caveolin and clustered GPI-anchored proteins normally, yet mis-sorted them, our results also indicate that clustering and caveolin are not sufficient for apical targeting, and that additional factors are required for the accurate apical sorting of GPI-anchored proteins.