A maximum likelihood streak detection (MLSD) algorithm is presented for the slow moving target(target velocity in the focal plane array is smaller than 1 pixel per frame)in deep space,it can implement the detection of slow target under the low Signal to Noise Ratio (SNR) condition effectively. The target pulse shape information is included in the signal model,and it is an improved version of the Maximum Value Projection Detection (MVPD) algorithm. The image signal mathematics model is built based on gaussian noise distribution,Its theory detection performance model and the real-time performance are analyzed. Monte Carlo simulation technique is adopted in comparing and analyzing the detection performance of MLSD and MVPD algorithm using simulated sequence images. The result shows that the detection probability of the MLSD reaches to 95 percent when the input SNR is 3.5,and the input SNR required for the MLSD is 2.5 lower than MVPD under the same detection probability. (required an input SNR of 6 for the MVPD). The real-time processing capability of the MLSD can obtain 31.25 Mb/s.