The invention concerns a classifying apparatus, used in particular for recognising or characterising odours, applying a combination of statistical methods and neuronal networks to classify into the appropriate class instances of a plurality of different classes presented to the apparatus. The apparatus comprises a processing unit for determining for each class (j) a subspace (SEj) wherein the instances of said class are best separated from instances of other classes, said subspace being defined by synthetic variables (VDj), and for determining a discriminating subspace (SED) defined by the whole set of synthetic variables identified for the whole set of classes. Each neuron in the neuronal input layer corresponds to one of the variables defining the discriminating space (SED) and each neuron of the output layer corresponds to one of the classes.