This work presents a gas discrimination approachfor Open Sampling Systems (OSS), composed of non-specificmetal oxide sensors only. In an OSS, as used on robots or insensor networks, the sensors are exposed to the dynamics of theenvironment and thus, most of the data corresponds to highlydiluted samples while high concentrations are sparse. In addition,a positive correlation between class separability and concentra-tion level can be observed. The proposed approach computes theclass posteriors by coupling the pairwise probabilities betweenthe compounds to a confidence model based on an estimation ofthe concentration. In this way a rejection posterior, analogous tothe detection limit of the human nose, is learned. Evaluation wasconducted in indoor and outdoor sites, with an OSS equippedrobot, in the presence of two gases. The results show that theproposed approach achieves a high classification performancewith a low sensitivity to the selection of meta parameters.