Abstract: A feasible approach to the practical application of radar automatic target recognition(RATR) based on high resolution range profile(HRRP) is enrolling data while learning and modeling interactively and concurrently.On the assumption that echoes in range cells are independent Gaussian distributed,online learning parameters are first deduced,and then a two-threshold method based on probabilistic density fanction is proposed.The threshold S_A picks the outliers and the threshold S_B reduces the mismatch between the statistics model and the online real HRRP data.Experiments based on measured data show the effectiveness of the proposed method.