Abstract: A radar classifies an unknown target illuminated with a large bandwidth pulse. The large bandwidth pulse may be algorithmically synthesized. The target reflects the large bandwidth pulse to form a return. The return is digitized into digital samples at range bin intervals. A computer extracts unknown range and amplitude pairs descriptive of the unknown target from the digital samples. Some range and amplitude pairs are located within one range bin interval. Principle scatterers are extracted from the unknown range and amplitude pairs using Modified Forward backward linear Prediction to form an unknown feature vector for the target. A plurality of pre-stored, known feature vectors containing known range and amplitude pairs are retrieved from the computer. The known range and amplitude pairs are descriptive of known targets, and are grouped in clusters having least dispersion for each of the known targets.