In this report we describe a novel, bipartite vesicular stomatitis virus (VSV) replication system which was used to study the effect of mutations in the transcription start sequence on transcript initiation and 5-mRNA modifications. The bipartite replication system consisted of two genomic RNAs, one of which (VSV⌬G) was a recombinant VSV genome with the G gene deleted and the other (GFC) contained the G gene and two non-VSV reporter genes (green fluorescent protein [GFP] and chloramphenicol acetyltransferase [CAT]). Coinfection of cells with these two components resulted in high-level virus production and gave titers similar to that from wild-type-VSV-infected cells. Mutations were introduced within the first 3 nucleotides of the transcription start sequence of the third gene (CAT) of GFC. The effects of these changes on the synthesis and accumulation of CAT transcripts during in vivo transcription (e.g., in infected cells), and during in vitro transcription were determined. As we had reported previously (E. A. Stillman and M. A. Whitt, J. Virol. 71:2127-2137, 1997), changing the first and third nucleotides (NT-1 and NT-3) reduced CAT transcript levels in vivo to near undetectable levels. Similarly, changing NT-2 to a purine also resulted in the detection of very small amounts of CAT mRNA from infected cells. In contrast to the results in vivo, the NT-1C mutant and all of the second-position mutants produced near-wild-type amounts of CAT mRNA in the in vitro system, indicating that the mutations did not prevent transcript initiation per se but, rather, generated transcripts that were unstable in vivo. Oligo (dT) selection and Northern blot analysis revealed that the transcripts produced from these mutants did not contain a poly(A) ؉ tail and were truncated, ranging in size from 40 to 200 nucleotides. Immunoprecipitation analysis of cDNA-RNA hybrids with an antibody that recognizes trimethylguanosine revealed that the truncated mutant transcripts were not properly modified at the 5 end, indicating the transcripts either were not capped or were not methylated. This is the first demonstration that transcript initiation and capping/methylation are separable events during VSV transcription. A model is proposed in which polymerase processivity is linked to proper 5-end modification. The model suggests that a proofreading mechanism exists for VSV and possibly other nonsegmented minus-strand RNA viruses, whereby if some transcripts do not become capped during transcription in a normal infection, a signal is transduced such that the polymerase undergoes abortive elongation and the defective transcript is terminated prematurely and subsequently degraded.