ESS models for the evolution of seed size variation assume that seedlings compete with each other for the occupancy of 'safe sites' or vegetation gaps. If mortality rates are high and/or frequency-independent, ESS models reduce to the classical model of Smith and Fretwell which predicts that a single, optimum seed size should occur. We tested whether seedlings compete with one another by following the survival of seedlings colonizing experimental gaps in a grazed grassland community. In small gaps (3 cm diameter) the proximity of established plants slightly, though significantly reduced seedling survival, but density-dependent mortality also occurred among seedlings in these gaps. In larger gaps (6 cm, 9 cm diameter) survival was significantly positively frequency-dependent. These results strongly support the validity of ESS models.