A Lagrangian stochastic larva tracking model based on a 3-dimensional (3D) high resolution wind-driven coastal circulation model is used to study the dispersal of benthic larvae. The larva tracking model includes 3D advection and turbulence, and a species-specific larval swimming behaviour that accounts for ontogenic changes, sensitivity to light exposure and inter-individual variability. Larval mortality can also be included. The dispersal model is applied to Owenia fusiformis larvae, whose swimming behaviour description is based on both existing data and new complementary measurements. Larval velocities (resulting from both settling and swimming behaviour) were measured with actographic equipment and ranged between –1 and 0.9 mm s–1. Measured swimming activity rates were lower than 50%. The sensitivity study of larval dispersal in March–April 1999 showed that: (1) the dispersal of neutrally buoyant passive larvae is more sensitive to the physical forcings resolution, because of both advection and diffusion processes, than to the variability of spawning locations within neighbouring grid cells (up to 1 km apart) in Banyuls Bay (France, NW Mediterranean); and (2) a physical barrier, located at 20 m deep in Argelès (France, NW Mediterranean) and 30 m deep in Banyuls Bay, separated nearshore and offshore larval dispersal in 1999. The final positions and local retention of larvae released in Banyuls Bay and Argelès result from: (1) the balance between the 3D turbulence, larval settling velocity (~0.8 mm s–1) and swimming activity rate; and (2) natural mortality, although the effect is not proportional to survival rates. High resolution larvae dispersal patterns for O. fusiformis in Banyuls Bay suggest that self-recruitment was low in the Banyuls population during spring 1999 and confirm that post-settlement deposit patterns observed there in May 1999 were insignificant. In addition, interconnections between the Argelès and Banyuls populations can exist.