The classical land-cover classification methods based on remotely sensed data have often led to unsatisfactory results, partly due to their intrinsic limitations. In effect, parametric procedures such as the maximum-likelihood classifier are statistically stable and robust but lack in flexibility and in the capability of making correct area estimates. On the other hand, nonparametric classifiers are generally too sensitive to distribution anomalies and are critically dependent on training sample sizes. A solution to these problems is represented by the insertion of prior probabilities derived from a nonparametric process in a conventional parametric classifier. In the present paper an example of such a method is put forward in order to merge the advantages of parametric and nonparametric strategies without the relevant shortcomings.