An efficient method is proposed to track the motion of single molecules fast and accurately. Single molecule localization is realized by image de-noising and applying a centroid algorithm based on Gaussian mask. Then the corresponding locations are linked into trajectories taking into consideration the characteristic intensity and actual displacement between subsequent frames. The method is quantitatively evaluated by using simulated sequential images at variable signal-to-noise ratios (SNRs). Compared with two-dimensional direct Gaussian fitting which is considered to be the best technique for deriving centers of single molecules,the method proposed here not only shortens the computational time to 1/10 but is also superior over the whole range of SNRs yielding lower bias and standard deviation. Particularly,at low SNRs,where the two-dimensional direct Gaussian fitting doesn′t work properly,the center coordinates can still be reconstructed. A speed of 4-5 frame/s (one million pixels per image) has been achieved in the current study on a commercial personal computer.