Providing fast, practical, and reasonable estimates, pixel-level classifiers are commonly used for land-use classifications. However, their efficiency is generally poor for heterogeneous land classes for which spectral/thematic observations do not follow a normal probability distribution. This study introduces a statistical moment form classifier—the probabilistic distance classifier (PDC)—considering both spectral distances between considered pixels and supervised classes, and sample probabilities of the spectral/thematic observations. Calibrated with a representative land-use data set, the PDC reduces the classification process uncertainty arising from unknown probabilistic population behaviors of the supervised classes and thus improves the accuracy of the classification results. Considering five supervised land classes, the land-use map of Istanbul was produced with the PDC as well as with the pixel-level supervised classifiers most commonly used in the literature. The efficiency of the classifiers was evaluated based on error matrices via producer, user, and overall accuracies. Kappa analyses were applied to test the statistical significance of the classification results. This study showed better classification results with PDC than the other pixel-level classifiers considered, and the PDC proved to be a statistically significant classifier for land-use classification, especially in case of heterogeneous land classes with a non-normal probabilistic behavior.