A new steepest descent adaptive filter algorithm derived from a newly devised performance index function is presented. The performance function of the new algorithm is introduced from that of the least mean square (LMS) considering that the stochastic steepest descent method utilises a gradient search in order to minimise the performance function iteratively. Through mathematical analyses and computer simulations, it is verified that there are significant improvements in convergence speed and misadjustment error. Nevertheless its computational simplicity and robustness are maintained with little degradation (compared to those of the LMS algorithm). The new algorithm can be interpreted as a new kind of variable step size adaptive algorithm, and in this respect a modified method is proposed in order to reduce the noise caused by fluctuation of the varying step size.