In agricultural systems, polyphagous beetles and spiders are abundant components of the beneficial arthropod community. Although data on the dietary ranges of these groups is increasing, remarkably little is understood regarding how individuals interact with their prey at small spatial scales. We demonstrate the utility of a spatially-explicit network model that integrates predator behaviour using predator-prey co-occurrences. Three co-occurrence matrices, one each for June, July and August, were generated using Vortis suction sample data collected from an 80 point grid imposed on a field of winter wheat. Heuristic predator-prey linkages, based on positive spatial co-occurrence, were imposed on these three matrices to create networks. It was found that primary consumers were highly aggregated and showed a strong tendency to co-occur. This contrasted with patterns of predator–predator or predator–prey co-occurrences that either aggregated to their prey or were weak and more scattered. These patterns could not be explained by either competition for resources or body size differences. Procrustean methods indicated that the networks were temporally dynamic, consistently achieving rates of turnover >60%. A negative relationship was found between decreasing predator–prey co-occurrence in the network and the number of prey positives in the guts of those predators. For large polyphagous beetles, the closer they were to their prey at the field scale, the more likely they were to have eaten them. This simple underlying relationship suggests that spatial co-occurrence networks can be used to predict feeding behaviour and could make a valuable contribution to food web structuring.