The direct evidence of Radar Target Recognition is the backscatter energy and its distribution of objects, which is concluded by imaging process of Synthetic Aperture Radar (SAR) in the two-dimension image domain. So the issues about Automatic Target Recognition (ATR) of SAR are often a "post-process" following SAR imaging. A method of target detection in the state of non-imaging based on analyzing the features of all kinds of targets on SAR echo data is presented in this paper, where the targets are detected in Range-Doppler domain (RDD). Based on analysis of target features in RDD, an algorithm of target detection in a RDD image is developed. The experimental results indicate the validity of the algorithm for special kinds of target detection. The algorithm is simple, able to recurrent, easy for real-time processing and hardware implementation. It can also apply to getting target alarm with real-time SAR data for the pre-selected targets and do the imaging process selectively at the same time.