Theories of body size-based community ecology have several assumptions, i.e., the ubiquitous of size-specific predator-prey interactions via the ontogenetic niche shift of predators and these effects on interaction strength. I empirically tested these assumptions by describing a size-structured food web of a temperate stream. For 7 predatory fishes and 55 prey species, the body mass of individuals and ontogenetic niche shift of predators were examined. I constructed size-structured food web by separating the two carnivorous fishes (CFs) and prey on several size classes but not the five omnivorous fishes (OFs), because the CFs significantly increased prey individual size with growing while the OFs did not. Approximately 44% of all interactions was size-specific in the food web. A statistical model predicted the peak of the strength of interactions shifted to larger direction along a prey size gradient when CF individuals were large. In contrast, the OFs did not show that pattern, probably because of the constraint of a foraging trait. These results indicate the theoretical assumptions well explain CF–prey interactions, whereas OF–prey interactions may be explained by other mechanisms in the study site.