This paper is focused on the feature extraction techniques of radar high range resolution profiles (HRRPs). In order to release the translational sensitivity of HRRPs, two translation invariant features, the central moments and distribution entropy, are extracted from the HRRPs and combined to form a new feature vector. Experiment on real data of three airplanes in flight is implemented to evaluate the recognition performance of the combined feature, using the nearest neighbour (NN) classifier and the support vector machine (SVM) classifier, respectively. Experimental results demonstrate that the combined feature can significantly enhance the separability of different targets and improve the average recognition rate of HRRP target recognition.