The automatic aircraft recognition in aerial images has been a topic receiving the maximum attention in the literature for more than two decades due to its practical value. Since the original images may be obtained in various uncertain environment, defects such as noise/background clutter, skewness, partial occlusion and blue may affect the image quality which lead to high environment dependence of some existing aircraft detection systems/ algorithms. This paper proposes a novel approach based on the Maximum Likelihood Ratio Test (MLRT) to extract different aircraft from the real aerial images. Objects in the tested images are corrupted randomly by the various defects. Without any extra information, the successful detection rate of the aircraft reaches 96% and that of the nonplane objects reaches 100% when only six aircraft references are chosen. The experiments are carried out to test sensitivity of the proposed method to the noise and partial occlusion. The comparisons with the Moment Fourier Descriptors (MFD) are also given.