Herpes simplex virus infection of mammalian hosts involves lytic replication at a primary site, such as the cornea, translocation by axonal transport to sensory ganglia and replication, and latent infection at a secondary site, ganglionic neurons. The virus-encoded thymidine kinase, which is a target for antiviral drugs such as acyclovir, is not essential for lytic replication yet evidently is required at the secondary site for replication and some phase of latent infection. To determine the specific stage in viral pathogenesis at which this enzyme is required, we constructed virus deletion mutants that were acyclovir resistant and exhibited no detectable thymidine kinase activity. After corneal inoculation of mice, the mutants replicated to high titers in the eye but were severely impaired for acute replication in trigeminal ganglia and failed to reactivate from ganglia upon cocultivation with permissive cells. Nevertheless, latency-associated transcripts were expressed in neuronal nuclei of ganglia from mutant-infected mice and superinfection of the ganglia with a second virus rescued the latent mutant virus. Thus, contrary to a widely accepted hypothesis, the thymidine kinase-negative mutants established latent infections, implying that neither thymidine kinase activity nor ganglionic replication is necessary for establishment of latency. Rather, thymidine kinase appears to be necessary for reactivation from latency. These results suggest that acyclovir-resistant viruses could establish latent infections in clinical settings and have implications for the use of genetically engineered herpesviruses to deliver foreign genes to neurons.