Human exposure to xenobiotics is usually estimated by indirect methods. Biological monitoring has emerged during the last decade to improve assessment of exposure. However, biomonitoring is still an analytical challenge, because the amounts of sample available are often very small yet analysis must be as thorough and sensitive as possible. The purpose of this work was to develop an untargeted "exposomics" approach by using ultra-high-performance liquid chromatography coupled to high-resolution mass spectrometry (UHPLC-HRMS), which was applied to the characterization of pesticide metabolites in urine from pregnant women from a French epidemiological cohort. An upgradable list of pesticides commonly used on different crops, with their metabolites (more than 400 substances) was produced. Raw MS data were then processed to extract signals from these substances. Metabolites were identified by tandem mass spectrometry; putative identifications were validated by comparison with standards and metabolites generated by experiments on animals. Finally, signals of identified compounds were statistically analyzed by use of multivariate methods. This enabled discrimination of exposure groups, defined by indirect methods, on the basis of four metabolites from two fungicides (azoxystrobin, fenpropimorph) used in cereal production. This original approach applied to pesticide exposure can be extended to a variety of contaminant families for upstream evaluation of exposure from food and the environment.