Voltage-gated sodium channels are critical for neuronal activity, and highly intolerant to variation. Even mutations that cause subtle changes in the activity these channels are sufficient to cause devastating inherited neurological diseases, such as epilepsy and pain. However, these channels do vary in healthy tissue. Alternative splicing modifies sodium channels, but the functional relevance and adaptive significance of this splicing remain poorly understood. Here we use a conserved alternate exon encoding part of the first domain of sodium channels to compare how splicing modifies different channels, and to ask whether the functional consequences of this splicing have been preserved in different genes. Although the splicing event is highly conserved, one splice variant has been selectively removed from Nav1.1 in multiple mammalian species, suggesting that the functional variation in Nav1.1 is less well tolerated. We show for three human channels (Nav1.1, Nav1.2 and Nav1.7) that splicing modifies the return from inactivated to deactivated states, and the differences between splice variants are occluded by antiepileptic drugs that bind to and stabilize inactivated states. A model based on structural data can replicate these changes, and indicates that splicing may exploit a distinct role of the first domain to change channel availability, and that the first domain of all three sodium channels plays a role in determining the rate at which the inactivation domain dissociates. Taken together, our data suggest that the stability of inactivated states is under tight evolutionary control, but that in Nav1.1 faster recovery from inactivation is associated with negative selection in mammals.