Cell-surface proteins are associated with the lipid bilayer either as membrane-spanning molecules or as glycosyl phosphatidylinositol (GPtdIns)-linked proteins. Proteins destined for GPtdIns anchoring are synthesized as precursors with a hydrophobic C-terminal transmembrane domain, which is removed during the processing of these proteins in the endoplasmic reticulum (ref. 1). We have investigated the structural requirements for GPtdIns anchoring through the study of two closely related proteins which exhibit alternative membrane attachment. The IgG Fc receptor, Fc gamma RIII, is GPtdIns-linked on neurophils (III-1) whereas on natural killer (NK) cells and macrophages it is found as a transmembrane-anchored molecule (III-2), able to mediate antibody-dependent cellular cytotoxicity and phagocytosis. At the primary structural level, the III-1 gene differs from that encoding III-2 by only nine nucleotide substitutions, which result in six amino-acid differences, and the absence of 21 amino acids at the C terminus. We have analysed a series of III-1 and III-2 mutants in transient expression assays, and show that Ser 203 in the GPtdIns attachment domain is the dominant residue in determining whether the molecule can be GPtdIns-anchored. As in the case of its murine homologue, Fc gamma RII alpha, surface expression of the III-2 molecule is dependent on co-expression of a second subunit, the gamma chain of F epsilon RI. Our data also suggest that gamma chain can associate with the III-1 precursor, preventing GPtdIns attachment, favouring instead a transmembrane form.