in order to control the risks to which workers are exposed at their workstations, it is necessary to know the hazardous properties of the substances they use manually or in industrial processes. Most of these data are summarised in the safety data sheets of the handled substances and are usually obtained by experimental tests. In addition to these, INERIS develops predictive models of the hazardous physico-chemical properties of chemicals such as explosibility or flammability of nitrated compounds (Fayet, 2011), organic peroxides (Prana, 2014) or ionic liquids (Diallo, 2012). These models, known as QSPR for Quantitative Structure-Property Relationships, are based on correlations between the macroscopic property and descriptors of the molecular structure of a substance. In addition, quantum chemistry tools are used both for the definition of molecular descriptors taking into account the reactivity of substances but also to characterise the reactive mechanisms involved, for example, in the decomposition of nitrated compounds (Fayet, 2009). The QSPR methods have so far been mainly dedicated to toxicological properties but are increasingly used for physico-chemical properties (Dearden, 2013). The development of such models has been encouraged in particular in the context of the REACH Regulation on the basis of validation principles established by the OECD for the use of these models in a regulatory framework (OECD, 2007). Therefore, all models developed at INERIS are validated according to these validation principles so that they can be used in a regulatory framework, proposed to the JRC (Joint Research Center) to integrate their QSAR/QSPR model base and made available in existing tools (such as the ECHA/OECD QSAR Toolbox). For example, the QSAR Toolbox (http://www.qsartoolbox.org/), initially focused on toxicological properties, includes the first model dedicated to a dangerous physico-chemical property in its latest version made available in December 2014. This model, developed by INERIS, concerns the prediction of sensitivity to the impact of nitroaliphatic compounds (Prana, 2012). Predictions from these predictive models make it possible to anticipate the hazards of substances, upstream of testing and/or when testing is not feasible (e.g. for new substances still in the synthesis phase). In particular, these tools are highly relevant in screening procedures to select products with both desired functional properties and limited hazards for the worker. They will therefore be able to support the development of safer substances, processes and workstations (safety-by-design).