The harvester ant Messor barbarus can be responsible for high weed seed losses in dryland cereals in Spain. Because weeds occur in patches, harvester ants have to be able to find and exploit patches. However, seed patches can differ in size and may, therefore, differ in the probability of being discovered and exploited. Here, 90 patches varying in size from 0.25 to 9 m2 were created in three 50 x 50 m subareas in a cereal field. Oat seeds were sown, as weed seed surrogates, in the patches at 2000 seeds m-2. After 24 h, those remaining were collected and the exploitation rate (the percentage of seeds removed per patch discovered by ants) was estimated. Harvester ant nests and the location of the seed patches were georeferenced and used to estimate distances between them. The patch encounter rate (the proportion of patches discovered by the ants) decreased slightly, but significantly, with decreasing patch size, though not the exploitation rate, which was lowest in the smallest patches (78-94%) and highest in the largest (99-100%). Seed patches that were not found or partially exploited were mostly located in subareas with a lower ant nest density or a longer distance away from the nearest nest than seed patches that were fully exploited. The results of this study indicate that the interaction between the spatial distribution of ant nests and the patchy distribution of seeds can create opportunities for seeds to be subjected to lower levels of predation.