In nature, arthropod-borne viruses (arboviruses) perpetuate through alternating replication in vertebrate and invertebrate hosts. The trade-off hypothesis proposes that these viruses maintain adequate replicative fitness in two disparate hosts in exchange for superior fitness in one host. Releasing the virus from the constraints of a two-host cycle should thus facilitate adaptation to a single host. This theory has been addressed in a variety of systems, but remains poorly understood. We sought to determine the fitness implications of alternating host replication for West Nile virus (WNV) using an in vivo model system. Previously, WNV was serially or alternately passed 20 times in vivo in chicks or mosquitoes and resulting viruses were characterized genetically. In this study, these test viruses were competed in vivo in fitness assays against an unpassed marked reference virus. Fitness was assayed in chicks and in two important WNV vectors, Culex pipiens and Culex quinquefasciatus. Chick-specialized virus displayed clear fitness gains in chicks and in Cx. pipiens but not in Cx. quinquefasciatus. Cx. pipiens-specialized virus experienced reduced fitness in chicks and little change in either mosquito species. These data suggest that when fitness is measured in birds the trade-off hypothesis is supported; but in mosquitoes it is not. Overall, these results suggest that WNV evolution is driven by alternate cycles of genetic expansion in mosquitoes, where purifying selection is weak and genetic diversity generated, and restriction in birds, where purifying selection is strong.