Many chemical, physiological and trophic factors are known to be important in the bioaccumulation processes and trophic transfer of PCB in the biota. Understanding the primary factors influencing PCB contamination of fishes is critical for predicting and assessing risks to upper-trophic levels consumers including humans. We proposed here to (1) identify PCB contamination pathways that could explain between and within species variability in fish concentration levels; and (2) describe PCB transfer along fish food chain. Three freshwater river fishes (Barbel, Bream and Chub) were sampled in three sites along the Rhone river (France), where fish consumption is partially prohibited because of many exceeding of the European sanitary level. We showed that diet habitat exploitation is an essential factor, the sediment compartment playing a central role in the contamination process. By combining stable isotope mixing models and stepwise regression, it appeared that fish length, PCB concentration in the sediment and foraging habitat (exploitation of detrital carbon sources) explained around 80% of the within- and between species variability observed in PCB concentrations. The main important factor in fish PCB contamination is thus not what fishes eat but rather how contaminated the food they consume is, suggesting that spatial gradients of contamination are more important than the type of food consumed and its trophic status. A bioaccumulation food-web model, based on physiological processes, was then developed in order to describe PCB transfer along the food chain of these fish species. The use of Bayesian Inference to calibrate the functions involved allowed to pass on data variability and parameter uncertainty to model predictions and provided a credibility interval around them. Results concerning model predictions and sensibility analyses are presented and discussed in a risk assessment perspective. By linking sediment to fishes contamination, our model can help in determining sediment management guidelines in the future.