Models describing the evolution of dispersal strategies have mostly focused on the evolution of dispersal rates. Taking trees as a model for organisms with undirected, passive dispersal, we have developed an individual-based, spatially explicit simulation tool to investigate the evolution of the dispersal kernel, P(r), and its resulting cumulative seed-density distribution, D(r). Simulations were run on a variety of fractal landscapes differing in the fraction of suitable habitat and the spatial autocorrelation. Starting from a uniform D(r), evolution led to an increase in the fraction of seeds staying in the home cell, a reduction of the dispersal mortality (arrival in unsuitable habitat), and the evolution of 'fat-tailed' D(r) in autocorrelated landscapes and approximately uniform D(r) in random landscapes. The evolutionary process was characterized by long periods of stasis with a few bouts of rapid change in the dispersal rate.