The paper describes an algorithm to detect and classify electrical events related to power quality. The events detection is based on monitoring the statistical characteristics of the energy of the error signal, which is defined as the difference between the monitored waveform and a sinusoidal wave generated with the same magnitude, frequency, and phase as the fundamental sinusoidal component. The novel feature is event recognition based on a neural network that uses the error signal as input. Multi-rate techniques are also employed to improve the system operation for on-line applications. Software tests were performed showing the good performance of the system.