A pelagic larval stage is found in nearly all demersal marine teleost fishes, and it is during this pelagic stage that the geographic scale of dispersal is determined. Marine biologists have long made a simplifying assumption that behaviour of larvae—with the possible exception of vertical distribution—has negligible influence on larval dispersal. Because advection by currents can take place over huge scales during a pelagic larval stage that typically lasts for several days to several weeks, this simplifying assumption leads to the conclusion that populations of marine demersal fishes operate over, and are connected over, similar huge scales. This conclusion has major implications for our perception of how marine fish populations operate and for our management of them. Recent (and some older) behavioural research—reviewed here—reveals that for a substantial portion of the pelagic larval stage of perciform fishes, the simplifying assumption is invalid. Near settlement, and for a considerable portion of the pelagic stage prior to that, larvae of many fish species are capable of swimming at speeds faster than mean ambient currents over long periods, travelling tens of kilometres. Only the smallest larvae of perciform fishes swim in an energetically costly viscous hydrodynamic environment (i.e., low Reynolds number). Vertical distribution is under strong behavioural control from the time of hatching, if not before, and can have a decisive, if indirect, influence on dispersal trajectories. Larvae of some species avoid currents by occupying the epibenthic boundary layer. Larvae are able to swim directionally in the pelagic environment, with some species apparently orientating relative to the sun and others to settlement sites. These abilities develop relatively early, and ontogenetic changes in orientation are seemingly common. Larvae of some species can use sound to navigate, and others can use odour to find settlement habitat, at least over small scales. Other senses may also be important to orientation. Larvae are highly aware of their environment and of potential predators, and some school during the pelagic larval stage. Larvae are selective about where they settle at both meso and micro scales, and settlement is strongly influenced by interactions with resident fishes. Most of these behaviours are flexible; for example, swimming speeds and depth may vary among locations, and speed may vary with swimming direction. In direct tests, these behaviours result in dispersal different from that predicted by currents alone. Work with both tropical and temperate species shows that these behaviours begin to be significant relatively early in larval development, but much more needs to be learned about the ontogeny of behaviour and sensory abilities in larvae of marine fishes. As a preliminary rule of thumb, behaviour must be taken into account in considerations of dispersal after the preflexion stage, and vertical distribution behaviour can influence dispersal from hatching. Larvae of perciform fishes are close to being planktonic at the start of the pelagic period and are clearly nektonic at its end, and for a substantial period prior to that. All these things differ among species. Larvae of clupeiform, gadiform and pleuronectiform fishes may be less capable behaviourally than perciform fishes, but this remains to be confirmed. Clearly, these behaviours, along with hydrography, must be included in modelling dispersal and retention and may provide explanations for recent demonstrations of self-recruitment in marine fish populations. Current work is directed at understanding the ontogeny of the gradual transition from planktonic to nektonic behaviour. Although it is clear that larvae of perciform fishes have the ability to strongly influence their dispersal trajectories, it is less clear whether or how these abilities are applied.