Abstract: A novel symbol duration estimation for long code direct sequence spectrum spread (DS-SS) signal is proposed. The estimator consists of two steps. Firstly, a coarse estimation is obtained based on a higher-order autocorrelation statistics and maximizing corresponding autocorrelation coefficient. Secondly, based on the principle of differential pseudo-noise (PN) code despreading and second-order cyclic correlation function, a fine estimation is reached. The coarse estimation keeps robust under low signal-noise-ratio (SNR) and small sample size than the fine one, and given the latter a reliable estimation range, so that performance improvement is obtained for the latter. The differential PN despreading is exploited to eliminate the interference effect on the cyclic correlation function by the PN code. Compared with existing theory simulations show that the proposed method is irrelative to the PN codes and obtains significant performance improvement.