QRS Complex is the most striking waveform within electrocardiogram (ECG), which provides much information about the current state of the heart. This paper proposes a novel QRS complex detector using 3-layer artificial feedforward back propagation neural network. The training method is Levenberg-Marquardt back propagation. After baseline drifts elimination and lowpass filtering, the features of every sample in ECG signal are extracted. Then the features of samples are inputted to the neural network. The algorithm is tested by recordings of MIT-BIH Arrhythmia Database and the accuracy is 99.5%.