The protease activities responsible for the cotranslational processing of the Semliki Forest virus structural polyprotein were investigated by using an in vitro transcription-translation system. Three cleavages released the individual chains from the nascent polyprotein in the order capsid, p62, 6K (a nonstructural peptide), and E1. We showed directly that the protease activity responsible for the release of the capsid protein resides in the capsid itself: by progressive truncation of the cDNA used for the SP6 transcription, we showed that a precursor containing as few as 38 residues of the p62 protein left at the C terminus of the capsid was still very efficiently cleaved in vitro. We further tested the possibility that serine-219 of the capsid is involved in autoproteolysis by site-directed in vitro mutagenesis. A change in the sequence Gly-Asp-Ser(219)-Gly, a tetrapeptide conserved among several animal serine proteases, to Gly-Asp-Arg-Ser-Thr was shown to completely abolish in vitro cleavage. This supports the notion that the capsid is a serine protease. The role of the capsid protease in the processing of the 6K junctions was then investigated by translations of a hybrid polyprotein in which the capsid and most of the p62 sequences are replaced by those of the secretory protein lysozyme. The cleavages and concomitant appearance of the 6K peptide occurred efficiently and were shown to require the presence of membranes. This demonstrates that the capsid protease is not required for those cleavages and suggests that a membrane-associated host protease is responsible for the cleavage.