Seed predation is an important component of seed mortality of weeds in agro-ecosystems, but the agronomic use and management of this natural weed suppression is hampered by a lack of insight in the underlying ecological processes. In this paper, we investigate whether and how spatial and temporal variation in activity-density of granivorous ground beetles (Coleoptera: Carabidae) results in a corresponding pattern of seed predation. Activity-density of carabids was measured by using pitfall traps in two organic winter wheat fields from March to July 2004. Predation of seeds (Capsella bursa-pastoris, Lamium amplexicaule, Poa annua and Stellaria media) was assessed using seed cards at the same sites and times. As measured by pitfall traps, carabids were the dominant group of insects that had access to the seed cards. In the field, predation of the four different species of seed was in the order: C. bursa-pastoris>P. annua>S. media>L. amplexicaule; and this order of preference was confirmed in the laboratory using the dominant species of carabid. On average, seed predation was higher in the field interior compared to the edge, whereas catches of carabids were highest near the edge. Weeks with elevated seed predation did not concur with high activity-density of carabids. Thus, patterns of spatial and temporal variation in seed predation were not matched by similar patterns in the abundance of granivorous carabid beetles. The lack of correspondence is ascribed to effects of confounding factors, such as weather, the background density of seeds, the composition of the carabid community, and the phenology and physiological state of the beetles. Our results show that differences in seed loss among weed species may be predicted from laboratory trials on preference. However, predator activity-density, as measured in pitfall traps, is an insufficient predictor of seed predation over time and space within a field.