To investigate the organization of Golgi glycosyltransferases and their mechanism of localization, we have compared the properties of a number of medial and late acting Golgi enzymes. The medial Golgi enzymes, N-acetylglucosaminyltransferase I and II (GnTI and GnTII) required high salt for solubilization and migrated as high molecular weight complexes on sucrose density gradients. In contrast, the late acting Golgi enzymes, beta1,4-galactosyltransferase and alpha1, 2-fucosyltransferase, were readily solubilized in low salt and migrated as monomers/dimers by sucrose density gradient centrifugation. Analysis of membrane-bound GnTI chimeras indicates that the formation of high molecular weight complexes does not require the transmembrane domain and cytoplasmic tail sequences of GnTI. Furthermore, a soluble form of GnTI, containing the stem region and catalytic domain, accumulated in the Golgi prior to secretion, in contrast to beta1,4-galactosyltransferase. Soluble GnTI, which also associated with high molecular weight complexes, was comparable with membrane-bound GnTI in its ability to glycosylate newly synthesized glycoproteins in vivo. Mutation of charged residues within the stem region of GnTI, known to be important for "kin recognition", had no effect on the efficiency of Golgi localization, the inclusion into high molecular weight complexes, nor functional activity in vivo. The differences in behavior between the medial and late acting Golgi enzymes may contribute to their differential localization and their ability to glycosylate efficiently in the correct Golgi subcompartment.