The substrate specificity of beta 1,4-N-acetylgalactosaminyltransferase has been analyzed using a fusion enzyme which consisted of the catalytic domain of the enzyme and the IgG binding domain of protein A, and also by extracts from cDNA transfectants, Both enzyme sources were capable of producing not only G(M2) and G(D2), but also asialo-G(M2), GalNAc-sialylparagloboside, and Gal-NAc-G(D1a) from appropriate accepters, although the efficiencies were at most 1-3% of those of G(M2)/G(D2). The biological significance of these low specificities was studied with transient and stable transfectant cells. From the results of transient expression of the cDNA, asialo-G(M2) expression appeared to inversely correlate with G(M3) synthase levels in those lines. Consequently, G(M2) seemed to be preferentially synthesized when both G(M3) and lactosylceramide are available, and asialo-G(M2) is synthesized in the absence of G(M3) synthesis. However, the results of double immunostaining of CHO transfectants with anti-G(M2) and anti-asialo-G(M2) antibodies indicated that another factor may be involved in asialo-G(M2) synthesis. From the in vitro assay using mixed acceptors, it was concluded that the presence of certain levels of G(M2) might enhance the asialo-G(M2) synthesis. These results suggest that even accepters showing low efficiencies in vitro might be used in certain cells depending on the availability of precursors, expression levels of other gangliosides, as well as the kinetic properties of tile enzyme, and the compartmentation of the glycosylation machineries in the cells.