Among lots of exigencies, two major requirements are necessary to assure the best performance of untargeted metabolomics-based approaches: (1) covering the maximum number of compounds present in the studied samples, and (2) determining the “optimal” analytical conditions to do so. The present work suggests an approach to address these exigencies. This approach is based on LC-HRMS untargeted analyses and metabolomics computational tools. It will be developed and applied to assess the optimal extraction protocol dedicated to analyze a wide range of pesticides formulates and microbial metabolites at once, all hidden in a complex environmental matrix: agricultural soil. Therefore, to compare five different extraction protocols applied on two types of soil and two formulated herbicides, four criteria were selected: (1) the coverage of compounds in term of polarity, (2) the quantitative performance, (3) the repeatability, and (4) the capacity to discriminate between contaminated and non-contaminated soils. To assess each criterion, different data analysis and visualization tools were used (e.g. hierarchically-clustered and polarity-segmented Heatmaps, van Krevelen diagrams, Euclidean Distances, RSD density plots, and OPLS-DA). They will be presented, explained and discussed in order to show their advantages, limitations and complementarities, as well as giving the best practices and tricks to render their exploration optimal. These tools finally allowed for the selection of the best extraction protocol. They are thus suggested to assess the performance of analytical methods when it comes to untargeted metabolomics analyses.