Vegetables grown on contaminated soils can have high levels of potentially toxic elements (PTE), which can lead to health risks for those who consume them. Such a situation is encountered in Wallonia (South Belgium) where former industrial activities have substantially increased PTE topsoil content. Vegetables like carrots and chards frequently exceed European quality standards (for Cd and Pb; EC 1881/2006). Therefore, predicting PTE concentrations in vegetables based on soil properties is useful to anticipate the risk of contamination of crops, as well as for specifying population exposure to PTE in a health risk assessment model. Here we outline the methodology we developed for the prediction of ten PTE (As, Cd, Cr, Cu, Hg, Mn, Mo, Ni, Pb and Zn) in plants, and we present the main results and the lessons that can be drawn. A dataset of 1718 pairs of soil and plant analyses was created from relevant datasets based on identical measurement protocols (PTE: nitric acid dissolution for plants, aqua regia extraction for soils). Soil-to-plant transfer models were estimated by a robust regression method for the 14 vegetables with at least 20 records in the database. Depending on the element, we found between 0 (Hg) and 10 (Zn) statistically significant prediction models. When retained in the model, organic matter and iron tend to decrease the ETP content of vegetables. As expected, high cation exchange capacity results in low soil PTE bioavailability. In general, we found predictive models for vegetables with the highest PTE contents such as lettuce. The PTE contents of fruit vegetables (tomatoes, zucchini) are generally low and infrequently predictable by models. Similarly, As, Cr and Mo content is generally more difficult to predict as the vegetable content was close to the analytical quantification limits. The results of this study are currently being used in a web tool, called SANISOL (http://sanisol.wallonie.be/), which provides recommendations for users of trace metal contaminated vegetable gardens in Wallonia.