The observations of natural olfaction led to the evidence that the processing of olfactory receptor signals contributes to properties such as performance stability, concentration invariance, and background suppression. Then, the study of olfactory processing has been regarded as a valuable source of inspiration for sensor arrays. In this work, a neural network architecture, introduced in the past as a model of the olfactory bulb, is used as a pre-processing step to analyse the data of an array of chemical sensors. Surprisingly, besides to enforce gas recognition, the network shows an interesting property of drift rejection. The potentialities of the network are demonstrated with data collected in a long-time experiment. The results provide strong evidence that bio-inspired processing of chemical sensors data actively improve the overall performance.