Coronaviruses contain an unusually long (27-32,000 ribonucleotide) positive sense RNA genome that is polyadenylated at the 3' end and capped at the 5' end. In addition to the genome, infected cells contain subgenomic mRNAs that form a 3' co-terminal nested set with the genome. In addition to their common 3' ends, the genome and the subgenomic mRNAs contain an identical 5' leader sequence. The transcription mechanism that coronaviruses use to produce subgenomic mRNA is not known and has been the subject of speculation since sequencing of the subgenomic mRNAs showed they must arise by discontinuous transcription. The current model called leader-primed transcription has subgenomic mRNAs transcribed directly from genome-length negative strands. It was based on the failure to find in coronavirus infected cells subgenome-length negative strands or replication intermediates containing subgenome-length negative strands. Clearly, these structures exist in infected cells and are transcriptionally active. We proposed a new model for coronavirus transcription which we called 3' discontinuous extension of negative strands. This model predicts that subgenome-length negative strands would be derived directly by transcription using the genome RNA as a template. The subgenome-length templates would contain the common 5' leader sequence and serve as templates for the production of subgenomic mRNAs. Our findings include showing that: 1. Replication intermediates (RIs) containing subgenome-length RNA exist in infected cells and are separable from RIs with genome-length templates. The RFs with subgenome-length templates are not derived by RNase treatment of RIs with genome-length templates. 2. The subgenome-length negative strands are formed early in infection when RIs are accumulating and the rate of viral RNA synthesis is increasing exponentially. 3. Subgenome-length negative strands contain at their 3' ends a complementary copy of the 72 nucleotide leader RNA that is found in the genome only at their 5' end. 4. RIs with subgenomic templates serve immediately as templates for transcription of subgenomic mRNAs. Because subgenomic mRNAs are not replicated, i.e., copied into negative strands that in turn are used as templates for subgenomic mRNA synthesis, we propose that the subgenome-length negative strands must arise directly by transcription of the genome and acquire their common 3' anti-leader sequence after polymerase jumping from the intergenic regions to the leader sequence at the 5' end of the genome. This would make negative strand synthesis discontinuous and subgenomic mRNA synthesis continuous, which is the opposite of what was proposed in the leader primed model.