The need for odor measurement and pollution source identification in various sectors (aeronautic, automobile, healthcare...) has increased in the last decade. Multisensor modules, such as electronic noses, seem to be a promising and inexpensive alternative to traditional sensors that were only sensitive to one gas at a time. However, the selectivity, the non-repetitiveness of their manufacture, and their drift remain major obstacles to the use of electronic noses. In this first work, we show how the mathematical modeling of the sensor response can be used to find new selectivity characteristics, different from those classically used in the literature. We identified new specific characteristics that have no physical meaning that can be used to find criteria for the presence of formaldehyde and nitrogen dioxyde alone or in a mixture. We discuss the limitations of the methodology presented and suggest avenues for improvement, with more precise modeling techniques involving symbolic regression.