1. Current formulations of functional responses assume that the prey is homogeneous and independent of intraspecific processes. Most prey populations consist of different coexisting size classes that often engage in asymmetrical intraspecific interactions, including cannibalism, which can lead to nonlinear interaction effects. This may be important as the size structure with the prey could alter the overall density-dependent predation rates. 2. In a field experiment with damselfly and dragonfly larvae, 16 treatments manipulated the density of a small prey stage, the presence of large conspecific prey and the presence of heterospecific predators. 3. Size structure in the prey (i.e. when both prey stages were present) decreased the impact of the predator on overall prey mortality by 25-48% at mid and high prey densities, possibly due to density-dependent size-structured cannibalism in the prey. The predation rates on small prey stages were determined by the interaction of large prey and predators. Predation rates increased with prey density in the absence of large prey, but predation rates were constant across densities when large conspecifics were present. 4. The functional response for unstructured prey followed a Holling type III model, but the predation rate for size-structured prey was completely different and followed a complex pattern that could not be explained with any standard functional response. 5. Using additional laboratory experiments, a mortality model was developed and parameterized. It showed that the overall prey mortality of size-structured prey can be adequately predicted with a composite functional response model that modelled the individual functional responses of each prey stage separately and accounted for their cannibalistic interaction. 6. Thus, treating a prey population as a homogeneous entity will lead to erroneous predictions in most real-world food webs. However, if we account for the effects of size structure and the intraspecific interactions on functional responses by treating size classes as different functional groups, it is possible to reliably predict the dynamics of size-structured predator-prey systems.