We analyzed a region of the capsid of canine parvovirus (CPV) which determines the ability of the virus to infect canine cells. This region is distinct from those previously shown to determine the canine host range differences between CPV and feline panleukopenia virus. It lies on a ridge of the threefold spike of the capsid and is comprised of five interacting loops from three capsid protein monomers. We analyzed 12 mutants of CPV which contained amino acid changes in two adjacent loops exposed on the surface of this region. Nine mutants infected and grew in feline cells but were restricted in replication in one or the other of two canine cell lines tested. Three other mutants whose genomes contain mutations which affect one probable interchain bond were nonviable and could not be propagated in either canine or feline cells, although the VP1 and VP2 proteins from those mutants produced empty capsids when expressed from a plasmid vector. Although wild-type and mutant capsids bound to canine and feline cells in similar amounts, infection or viral DNA replication was greatly reduced after inoculation of canine cells with most of the mutants. The viral genomes of two host range-restricted mutants and two nonviable mutants replicated to wild-type levels in both feline and canine cells upon transfection with plasmid clones. The capsids of wild-type CPV and two mutants were similar in susceptibility to heat inactivation, but one of those mutants and one other were more stable against urea denaturation. Most mutations in this structural region altered the ability of monoclonal antibodies to recognize epitopes within a major neutralizing antigenic site, and that site could be subdivided into a number of distinct epitopes. These results argue that a specific structure of this region is required for CPV to retain its canine host range.