Untargeted toxicological screening is an analytical challenge, given the high number of molecules and metabolites to be detected and the constant appearance of new psychoactive substances (NPS). The combination of liquid chromatography with high-resolution tandem mass spectrometry (HRMS/MS) in a data-dependent acquisition mode generates a large volume of high quality spectral data. Commercial software for processing MS data acquired during untargeted screening experiments usually compare measured features (mass, retention time, and fragmentation spectra) against a predefined list of analytes. However, there is a lack of tools for visualizing and organizing MS data of unknown compounds. Here, we applied molecular networking to untargeted toxicological screening. This bioinformatic tool allows the exploration and organization of MS/MS data without prior knowledge of the sample's chemical composition. The organization of spectral data is based on spectral similarity.Hence, important information can be obtained even before the annotation step. The link established between molecules enables the propagation of structural information.We applied this approach to three clinical and forensic cases with various matrices: (a) blood and a syringe content in a forensic case of death by self-injection, (b) hair segments in a case of drug-facilitated assault, and (c) urine and blood samples in a case of 3-methoxyphencyclidine intoxication. Data preprocessing with MZmine allows sample-to-sample comparison and generation of multisample molecular networks.Our present study shows that molecular networking can be a useful complement to conventional approaches for untargeted screening interpretation, for example for xenobiotics identification or NPS metabolism elucidation.