ABSTRACT: Understanding the response of predator populations to varying prey fields is a prerequisite for understanding prey population dynamics and to correctly parameterise multi-species stock assessment or ecosystem models. Previous analyses on the large scale feeding response of predator populations, however, came to unrealistic results for the North Sea. The observed feeding response types (e.g. negative prey switching) would lead to the extinction of prey populations when these become scarce. We analysed the large scale response of North Sea cod Gadus morhua and whiting Merlangius merlangus populations to varying prey fields using Generalised Additive Models (GAMs). Thereby, we took changes in predator–prey overlap explicitly into account, in contrast to previous analyses. The composition of the prey field and changes in predator–prey overlap had significant effects on the diet composition in the final GAM, explaining 65.6% of the variance. The existence of a large scale prey refuge at low prey abundances as proposed by the Holling type III functional response could be demonstrated from field data. The refuge was caused by active prey-switching behaviour of the predators, and also by a passive change in the availability of prey due to changes in predator–prey overlap associated with changes in the prey abundance. In addition, a rapid increase in relative stomach contents was observed if the prey populations passed the abundance threshold of the prey refuge leading to a predator pit. At even higher abundances a saturation effect in relative stomach contents was detected. This study demonstrates that current diet selection models can be significantly improved by taking into account changes in spatial predator–prey overlap.