1. Changes in the density of an organism near a boundary elements (edge effects) are often thought to be generated by changes in movement behaviour, but in most cases the mechanism underlying these effects is unknown. 2. We quantified the movement behaviour of a minute parasitic wasp, Anagrus columbi, in relation to edges in its habitat. This wasp attacks eggs of the planthopper Prokelisia crocea, which inhabits a wet prairie ecosystem composed of patches of its host plant prairie cordgrass (Spartina pectinata) interspersed within a matrix of mudflat, smooth brome (Bromus inermis) and native grasses. Two edge types are common in this system, cordgrass-mudflat and cordgrass-brome. 3. We conducted mark-recapture experiments in which wasps were released at the cordgrass-matrix edge and 50 cm within cordgrass or matrix, for both edge types. The marked wasps were recaptured using a grid of sticky traps. We fitted an advection-diffusion model to these data, yielding estimates of the diffusion rate and advection coefficient for cordgrass and matrix, for each release position and edge type. 4. The spatial distribution of wasps was well-described by the advection-diffusion model. The pattern suggests that marked wasps strongly biased their movements towards the edge when released in matrix, and to a lesser extent when released in cordgrass, while edge releases showed little bias. The advection coefficients were similar for the two edge types, as were the diffusion rates for the three substrates (cordgrass, brome, mudflat). The diffusive and advective components of movement were of comparable magnitude for matrix and cordgrass releases, suggesting equal amounts of directed and random movement. 5. Our results suggest the wasps are attracted to cordgrass patches across short distances, and that bias in their movements may concentrate them at the patch edge. Their edge behaviour is qualitatively different from that of the host insect. 6. The methodology described here could be readily adapted to other systems, where direct observations of movement are difficult but mark-recapture studies are feasible.