The first step to evaluate the effects of contamination on organisms is to study toxicokinetics. The bioaccumulation of contaminants by aquatic species is a variable phenomenon, since it depends on the characteristics of the environment, the properties of the contaminants and the species. Different toxicokinetic models have been developed to describe the accumulation of contaminants in aquatic food webs. In these models, the organism is often considered as a single compartment: the bioaccumulation is then described as the balance between uptake and elimination processes. The absorption process can involve both dissolved or trophic route. The diet of aquatic organisms is known to be an important route of bioaccumulation of contaminants. The elimination process includes excretion, metabolism and dilution by growth. To date, there are few models focusing on persistent organic contaminants. Furthermore, estimating models' parameters is generally done through a frequentist approach in two steps: first by estimating parameter(s) related to depuration and then estimating parameter(s) related to accucumulation. The problem by doing this is that depuration during the accumulation phase is neglected, while this process occurs in the two phases. The aim of our study is to propose a Bayesian framework to estimate the parameters of a biodynamic model together by considering simultaneously accumulation and depuration data. The posterior distribution obtained for all parameter will enable a more accurate assessment of bioaccumulation uncertainty. We illustrate our approach with the freshwater benthic invertebrate Gammarus fossarum exposed for 7 days to a sediment spiked with PCB153 and transferred to a clean media for 7 more days. The PCB153 concentrations in Gammarus fossarum increased from an initial concentration of 0.32 to 12.36 ng.g-1 ww (wet weight) at the end of accumulation step. When gammarids were transferred into a clean media, the PCB153 concentration in organisms decreased to 6.41 ng.g-1 ww at day 14. The bioaccumulation model assuming first-order kinetics was fitted to the data using Bayesian inference. The inference process quickly converged and thin posterior distributions were obtained for each parameter, meaning that data brough enough information to estimate preciselly each parameter. The median model predictions and their 95% credibility intervals showed a good fit of the model to the data.