Detection of explosive residues using portable devices for locating landmine and terrorist weapons must sat- isfy the application criteria of high reproducibility, specificity, sensitivity and fast response time. Vibrational spectroscopies such as Raman and infrared spectroscopies have demonstrated their potential to distinguish the members of the chemical family of more than 30 explosive materials. The characteristic chemical fingerprints in the spectra of these explosives stem from the unique bond structure of each compound. However, these spectroscopies, developed in the early sixties, suffer from a poor sensitivity. On the contrary, MEMS-based chemical sensors have shown to have very high sensitivity lowering the detection limit down to less than 1 picogram, (namely 10 part per trillion) using sensor platforms based on microcantilevers, plasmonics, or surface acoustic waves. The minimum amount of molecules that can be detected depends actually on the transducer size. The selectivity in MEMS sensors is usually realized using chemical modification of the active surface. However, the lack of sufficiently selective receptors that can be immobilized on MEMS sensors remains one of the most critical issues. Microcantilever based sensors offer an excellent opportunity to combine both the infrared photothermal spectroscopy in their static mode and the unique mass sensitivity in their dynamic mode. Optical sensors based on localized plasmon resonance can also take up the challenge of addressing the selectivity by monitoring the Surface Enhanced Raman spectrum down to few molecules. The operating conditions of these promising localized spectroscopies will be discussed in terms of reliability, compactness, data analysis and potential for mass deployment.