The spatial context in which seed predation occurs may modify the spatial structure of recruitment generated by seed dispersal. The Janzen-Connell (J-C) model predicts that granivores will exert greater pressure on the parent plant or at those sites where the density of dispersed seeds is higher. We have investigated how the probability of post-dispersal survival of Juglans australis varies with nut density across a hierarchy of spatial scales. We experimentally evaluated the survival of 3,120 nuts at three spatial scales: meso-scale (<or=1.5 ha), as forest sites with two densities of fruiting J. australis individuals; intermediate scale (<0.2 ha), as individual trees with two experimental crop sizes; small scale (<0.1 m(2)), as microsites with two factors (number of nuts and distance from source). Nut removal coincided with seed predation, a condition that allowed us to test the density-dependent seed predation hypothesis. We found that the probability of nut survival was greater at forest sites with higher J. australis density. Nut survival was not affected by nut density in the seed shadow of individual specimens: at sites where J. australis density was greater, the proportion of surviving nuts did not differ between microsites located at different distances from the parent plant, but it was greater at microsites with greater initial nut density. Nut survival depended on the scale at which rodents responded to nut density, being negatively density dependent at the meso-scale and spatially random at intermediate and small scales. At the meso-scale, excess nut supply increased the probability of nut survival, which is in agreement with a model of granivore satiation near the seed source. Rodent satiation at the meso-scale may favour maintenance of sites with high J. australis density, where individual trees may have greater probabilities of passing their genes onto the next stage of the dispersal cycle.