In the field of toxicological bioassays, the latest progress in Raman spectroscopy opens new research perspectives on a fast method of observing metabolic responses against toxic agents. This technique offers a multiparametric approach, providing an overview of the physiological changes that are caused by pollutants. However, physiological spectral fingerprints require complex chemometric methods for proper analysis. In this study, particular attention has been given to the elaboration of an "aberrant spectra" detection strategy to highlight the effects of arsenic on the bacteria Escherichia coli. This strategy significantly improved spectra classification, consistent with a dose-response effect of the four tested concentrations of the metal. Indeed, the correct classification score of the spectra increased from 88 to more than 99%. The exposure time effect has also been investigated. The fine analysis of Raman spectroscopy fingerprints enabled the design of different "spectral signatures", highlighting early and late effects of arsenic on bacteria. The observed variations are in agreement with the expected toxicity and encourage the use of Raman spectroscopy for toxic element detection.