The invention provides a training sequence variable step-size least-mean-square algorithm, in particular to a training sequence quick channel estimation method. Employing factors which are capable of adaptively tracking and controlling the changes of characteristic parameters through error estimation, the quick channel estimation method improves the convergence speed of the algorithm and acquires the weight coefficient of an optimum filter speedily. The invention discloses a training sequence least-mean-square characteristic parameter estimation method which is quicker in convergence speed and more precise than prior art. The method is easy to realize and is applicable in communication systems for modulated orthogonal frequency division multiplexing for channel estimation. Meanwhile, the idea of the invention can be applied to CDMA channel estimation devices and TDMA channel estimation devices, and can also be applied to LMS methods and the derivative methods thereof. The channel estimation method relates to the fields of communication, oil and seismic exploration, sonar, image processing, computer vision, biomedical engineering, vibration engineering, radar, remote control and telemetry, as well as aerospace field.