The relationship between predator and prey body sizes is an important property of food webs with potential implications for community dynamics and ecosystem functioning. To shed more light on this issue I here analyze the relationships between prey size, predator size and trophic position of consumers, using body size estimates of 697 species in 52 food webs. First I show that the relationship between predator and prey body sizes across many systems can be different from, and potentially obscure the true relationship within systems. More specifically, when data from all webs are aggregated average prey size is positively correlated to predator size with a regression slope less than unity, suggesting that predators become less similar in size to their average prey the larger the predator is, and consequently that the relative size difference between a predator and its prey should increase with the trophic position of the consumer. However, despite this I find the predator-prey body mass ratio to be negatively correlated to the trophic position of the consumer within many webs. The reason for this is that the across-webs pattern is not representative for the within-web relationship. Second, I show that the pattern observed is not compatible with a simple null-model for the distribution of trophic links between predators and prey. The observed relationship between predator size and mean prey size is for most webs significantly steeper than that predicted by the cascade model. Furthermore, the observed relationship also deviates significantly (but less so) from an ecologically more realistic model for the distribution of trophic links (the niche model). The results contradict the traditional Eltonian paradigm that predator-prey body mass ratios do not vary consistently across trophic levels. It is concluded that more studies are needed to establish the generality of the results and explore its dynamical implications.