Significance (120 mots max, doit être lisible par des étudiants qui ne sont pas dans le domaine) We present a demonstration by simulation of a non-destructive neutron interrogation system capable of identifying organic compounds in cargo containers. Unlike previous systems, which tried identifying organics molecules using only carbon, nitrogen and oxygen neutron-induced gamma rays, our approach allows determining the fraction of hydrogen although no hydrogen specific gamma ray can be measured. Since hydrogen manifests in the gamma ray spectrum in a complex way through changes in the amplitude of carbon, oxygen and nitrogen gamma ray peaks, we discovered that a neural network can cope with the problem complexity and allow interpreting the peaks amplitudes in term of hydrogen fraction. Such a system can improve the reliability of cargo containers screening by reducing false alarms. Abstract (250 mots max, compréhensible par tout scientifique) To inspect cargo containers, X-ray imaging can be complemented by fast neutron interrogation to provide indication concerning the nature of the transported goods. Organic goods are of special interest since they constitute a significant part of the merchandises. In addition, in the context of NRBC-E threats search in cargo containers, a nondestructive inspection system should also be able to detect explosives. Until now, fast neutron interrogation systems identify organic materials using characteristic neutron-induced gamma-ray peaks of carbon, oxygen and nitrogen. But identifying organics in this way can lead to ambiguities since no hydrogen peak can be measured with fast neutrons. However, it is known that hydrogen strongly modifies the neutron energy spectrum, which in turn affects the gamma-ray peaks amplitudes. The link between hydrogen fraction and gamma ray peaks amplitude being complex, no attempt has been performed to inverse this link until now. Simulations show however that a neural network that takes as inputs the heights of carbon, oxygen and nitrogen gamma ray peaks can indeed determine the hydrogen fraction. Simulations of realistic cases show that the use of a neural network indeed allows identifying compounds having similar fractions of carbon, oxygen and nitrogen but different hydrogen fraction, opening the way to more accurate materials identification.