A novel algorithm is presented for harmonic estimation in an active power filter. The current signal whose harmonics are to be estimated is modelled by using Fourier series. The fundamental frequency is estimated by using a non-linear least-squares approach and involves a one-dimensional search over the allowable frequency range in the power system. The estimated frequency is then used in the signal model to compute the harmonic magnitudes. The performance of a shunt APF using this approach for frequency and harmonic extraction has been studied by using both computer simulation and hardware implementation. It is shown that the active filter is able to track the changes in the load current waveform (both with respect to magnitude and waveshape) more accurately and faster than in the conventional methods. © 2009 The Institution of Engineering and Technology.