Abstract: A passive acoustical recognition technique of low altitude targets is studied in this paper. Based on the human auditory principles and the superiorities of feature learning, this technique combines the methods of Mel-scale spectrum extraction and sparse nonnegative matrix factorization. Firstly, the Mel-scale spectrums of different kinds of targets are taken as the feature matrices, their respective template matrices are obtained via learning with the sparse NMF tech-nique. Then, the grouped template matrices are used for decomposing the signal features of each set of training/testing data to obtain coding coefficients, which are regarded as the feature vectors in classification model. Finally, according to the characteristics of different targets, the multi-band feature extraction and sequential binary classification procedures are used for multi-target classification, and the classical MFCC features are treated as the baseline to compare with the results of the sparse NMF method. The classification experiments demonstrate that the sparse NMF method achieves significantly better results in both single-class target identification and multi-class target classification.