In this paper a novel synchronization method is proposed to obtain a reference signal for active power filters (APF), which is used to determine the fundamental component of the load current in-phase with the mains voltage, without any signal measurement for synchronization. The principle idea of this paper is based on the proposition that "the required minimum value of the mains current is the value at which the reactive power becomes zero". Therefore, the proposed reference signal extraction algorithm uses only the mains current. It is based on artificial neural networks (ANN) and when compared to the alternative methods, the proposed method comes forward with its simplicity in terms of hardware and software. In harmonic component extraction stage, an adaptive filter structure is used. Instead of common least mean squares (LMS) method, averaged least mean squares (A-LMS) method, which is generated by adding a moving average filter (MAF) block to the LMS algorithm, is preferred because it allows faster and more accurate determination of the fundamental harmonic component.