NAD:arginine mono-ADP-ribosyltransferases catalyze the transfer of ADP-ribose from NAD to the guanidino group of arginine on a target protein. Deduced amino acid sequences of one family (ART1) of mammalian ADP-ribosyltransferases, cloned from muscle and lymphocytes, show hydrophobic amino and carboxyl termini consistent with glycosylphosphatidylinositol (GPI)-anchored proteins. The proteins, overexpressed in mammalian cells transfected with the transferase cDNAs, are released from the cell surface with phosphatidylinositol-specific phospholipase C (PI-PLC), and display immunological and biochemical characteristics consistent with a cell surface, GPI-anchored protein. In contrast, the deduced amino acid sequence of a second family (ART5) of transferases, cloned from murine lymphoma cells and expressed in high abundance in testis, displays a hydrophobic amino terminus, consistent with a signal sequence, but lacks a hydrophobic signal sequence at its carboxyl terminus, suggesting that the protein is destined for export. Consistent with the surface localization of the GPI-linked transferases, multiple surface substrates have been identified in myotubes and activated lymphocytes, and, notably, include integrin α subunits. Similar to the bacterial toxin ADP-ribosyltransferases, the mammalian transferases contain the characteristic domains involved in NAD binding and ADP-ribose transfer, including a highly acidic region near the carboxy terminus, which, when disrupted by in vitro mutagenesis, results in a loss of enzymatic activity. The carboxyl half of the protein, synthesized as a fusion protein in E. coli, possessed NADase, but not ADP-ribosyltransferase activity. These findings are consistent with the existence at the carboxyl terminus of ART1 of a catalytically active domain, capable of hydrolyzing NAD, but not of transferring ADP-ribose to a guanidino acceptor.