A novel advanced harmonic detection method based on neural network (NN) is proposed in this paper. It is an adaptive harmonic detection method with variable step-size based on adaptive linear NN and self-adaptive noise countervailing principle. And this proposed method adopts a sliding integrator to extract the real tracing error and then uses a fuzzy adjuster with self-adjustable factor to modify the step-size. So the novel harmonic detection method can obtain fast convergence speed and high steady-state precision at the same time. Comparisons are made between conventional harmonic detection methods based on NN and the advanced method based on NN proposed in this paper. Finally detailed simulation and experimental results verify the validity and superiority of the advanced methods.