The use of laboratory animals for toxicity testing in chemical safety assessment meets increasing ethical, economic and legislative constraints. The development, validation and application of reliable alternatives for in vivo toxicity testing are therefore urgently needed. In order to use toxicity data obtained from in vitro assays for risk assessment, in vitro concentration-response data need to be translated into in vivo dose-response data that are needed to obtain points of departure for risk assessment, like a benchmark dose (BMD). In the present study, we translated in vitro concentration-response data of the retinoid all-trans-retinoic acid (ATRA), obtained in the differentiation assay of the embryonic stem cell test, into in vivo dose-response data using a physiologically based kinetic model for rat and human that is mainly based on kinetic model parameter values derived using in vitro techniques. The predicted in vivo dose-response data were used for BMD modeling, and the obtained BMDL10 values [lower limit of the 95 % confidence interval on the BMD at which a benchmark response equivalent to a 10 % effect size (BMR10) is reached (BMD10)] for rat were compared with BMDL10 values derived from in vivo developmental toxicity data in rats reported in the literature. The results show that the BMDL10 values from predicted dose-response data differ about sixfold from the BMDL10 values obtained from in vivo data, pointing at the feasibility of using a combined in vitro-in silico approach for defining a point of departure for toxicological risk assessment.