Manmade targets detection is a crucial function of synthetic aperture radar (SAR) automatic targets recognition (ATR) systems. At present, there exist sort of limited methods which are applicable to complex background. To treat this case, a method based on the intrinsic dimension of multiaspect angle image samples is proposed. This method doesn’t count on intensity of the image, whereas it estimates the intrinsic dimension of images samples of the region of interest (ROI) in high dimensional space, which are obtained from different aspect angles. The measurement of artificiality and naturalness is performed with the intrinsic dimension of ROI. In this way manmade targets are detected. The real data from Gotcha project are adopted to compare our method with that of fractal dimension to detect manmade targets. The experiment result shows that intrinsic dimension has a merit that it doesn’t count on image intensity and it performs better than the fractal dimension method in depicting the shape of targets.