Powerful data pretreatment strategies inspired from the field of metabolomics were adapted to chemical food safety context to enable samples discrimination by multivariate methods based on low abundance ions. A highly automated workflow was produced. The open-source XCMS package was used and efficient data filtration strategies were set up. Data were treated using Independent Components Analysis, and data mining strategies developed to automatically detect and annotate ions of low abundance by coupling blind data exploration strategies with a broad scale database approach. Our method was efficient in discriminating tea samples based on their contamination levels (even at 10 µg.kg-1) and detecting unexpected impurities in the spiking mix. Several "tracer" contaminants were considered, covering a broad range of physicochemical properties and structural diversity with overall 66% detected and annotated blindly. The methodology was successfully applied to a data set exhibiting only 3 "tracer" contaminants (at 50 µg.kg-1) and more product diversity.