We used the nuclear polyhedrosis virus of the gypsy moth, Lymantria dispar, to investigate whether the timing of transmission influences the evolution of virulence. In theory, early transmission should favour rapid replication and increase virulence, while late transmission should favour slower replication and reduce virulence. We tested this prediction by subjecting one set of 10 virus lineages to early transmission (Early viruses) and another set to late transmission (Late viruses). Each lineage of virus underwent nine cycles of transmission. Virulence assays on these lineages indicated that viruses transmitted early were significantly more lethal than those transmitted late. Increased exploitation of the host appears to come at a cost, however. While Early viruses initially produced more progeny, Late viruses were ultimately more productive over the entire duration of the infection. These results illustrate fitness trade-offs associated with the evolution of virulence and indicate that milder viruses can obtain a numerical advantage when mild and harmful strains tend to infect separate hosts.