The virus-specific components (nsP1-nsP4) of Semliki Forest virus RNA polymerase are synthesized as a large polyprotein (P1234), which is cleaved by a virus-encoded protease. Based on mutagenesis studies, nsP2 has been implicated as the protease moiety of P1234. Here, we show that purified nsP2 (799 amino acids) and its C-terminal domain Pro39 (amino acids 459-799) specifically process P1234 and its cleavage intermediates. Analysis of cleavage products of in vitro synthesized P12, P23, and P34 revealed cleavages at sites 1/2, 2/3, and 3/4. The cleavage regions of P1/2, P2/3, and P3/4 were expressed as thioredoxin fusion proteins (Trx12, Trx23, and Trx34), containing approximately 20 amino acids on each side of the cleavage sites. After exposure of these purified fusion proteins to nsP2 or Pro39, the reaction products were analyzed by SDS-polyacrylamide gel electrophoresis, mass spectrometry, and amino-terminal sequencing. The expected amino termini of nsP2, nsP3, and nsP4 were detected. The cleavage at 3/4 site was most efficient, whereas cleavage at 1/2 site required 5000-fold more of Pro39, and 2/3 site was almost resistant to cleavage. The activity of Pro39 was inhibited by N-ethylmaleimide, Zn(2+), and Cu(2+), but not by EDTA, phenylmethylsulfonyl fluoride, or pepstatin, in accordance with the thiol proteinase nature of nsP2.