Although very often cited in publications dealing with food products, electronic noses still pose many problems. One is the extraction of features from the response curve; in general, only the adsorption maximum is retained and input into a classification system. This paper describes a statistic-based methodology developed to extract the most pertinent features from the outputs of SnO2-gas sensor array. Several features are extracted from each sensor curve and also from its primary and secondary derivatives. They are then sorted, taking into account three specific indexes, designed to describe the repeatability, discrimination power and their redundancy. To generalise this approach, the acquisitions are carried out using various operating conditions. This protocol is applied to a set of model mixtures, representing wine with satisfactory and unsatisfactory tart or vinegar flavour. This paper shows that relevant information can be obtained from the curve maximum, but also from features related to derivatives. Moreover, the most efficient features are the same for the five sensors, which would seem to indicate that they should also be the most suitable ones for all SnO2 sensors.