To provide a tap coefficient training method and circuit, that can obtain tap coefficient with optimum stability, in which the occurrence of divergence or non-convergence of the tap coefficient will not occur, with respect to a characteristic change in a transmission line.

The tap coefficient training circuit consists of five tap coefficient arithmetic circuits 13a-13e, that receive a reception training signal y(D) from a transmission line and a transmission training signal X equivalent to a transmitter side training signal generated by a PRBS(pseudo random binary sequence) generator 11 and an encoder 12 and respectively output tap coefficients Ww(D) a-Ww(D)e and data signals Za-Ze which denotes the degree of adaptation of the tap coefficients extracted from an updated amount, by using LMS algorithms having different step sizes installed in the inside of the circuits and of a selection circuit 14 or the like, that receives the coefficients Ww(D)a-Ww(D)e and the data signals Za-Ze for selecting and outputting a Ww(D), whose converged value is minimum among the data signals Za-Ze. Moreover, the number of tap coefficient arithmetic circuits 13 is not limited to five but the required number of them may be provided.