In this paper, an Averaged-Least Mean Squares (A-LMS) based reference extraction method for active power filter (APF,) is proposed. With A-LMS, high convergence speed and accuracy are achieved while maintaining processing simplicity. Moving average filter (MAF), whose filter characteristics usually seen as insufficient for most applications, is an ideal structure for APF application. In APFs, the distorting signals occur at the multiple frequencies of the fundamental frequency component. If the side-lobes occurring in a MAF are adjusted to be at the harmonic frequencies then the reference signal extraction accuracy will be improved. Another advantage of A-LMS is its fast convergence. The adaptation constant, which is effective on convergence speed, cannot be chosen higher in LMS derived algorithms due to its noise increasing effect. On the other hand, in A-LMS algorithm, since the signal to noise ratio (SNR) can be kept minimum the adaptation constant can be chosen higher to increase the convergence speed. In this paper, after giving information about A-LMS structure, it is shown that A-LMS is an ideal reference signal extraction method for APF applications through simulation and practical test results. Copyright (c) 2012 Praise Worthy Prize S.r.l. - All rights reserved.