Food safety is threatened by the excessive accumulation by crops, of toxic trace elements (TEs) naturally occurring in agricultural soils. Among these contaminants, cadmium (Cd) is an important issue because for this harmful metal, the excessive exposure of humans through food products has been pointed out by health authorities. Therefore, the regulation limits for Cd in food products have been recently revised to be more strict (EC1323/2021) and different sectors have already brought up alerts indicating the non-compliance of some harvests with the new limits. Apart Cd, lead (Pb) has also been recently regulated more strictly (EC1317/2021) and Ni might also be regulated, based on a putative significant risk for health, in particular for children. We present a statistical modelling approach to predict the risk for a food crop to not comply with the European regulatory limits for regulated TEs. We initially concentrated on Cd in durum wheat but the approach is also extended to different other crops and contaminants. The modelling uses machine learning methods to predict a probability of non-compliance. The number of input variables are limited and they are easily measurable in the field for the practical use by the stakeholders. The models were built from databases derived from research projects and from data from literature. The model performances are good since for Cd and durum wheat, for example, based on cross-validation, around 80% of non-compliant samples are detected. For durum wheat, the model is currently freely available as a web service (https://ispa.bordeaux.inra.fr/services/blesur/) and it can readily be used by the agricultural sector. Extending this work to other crops and contaminants is under progress.