This paper proposes a novel algorithm for removing the artifacts of electrocardiogram (ECG) signal and extracting the QRS complex based on empirical mode decomposition (EMD). The method can remove both the noise from the power line interference and the baseline wander (BW) with the minimum distortion. R peaks can be exactly detected, and the width of QRS complex can be measured. Experiments are based on records from MIT-BIH arrhythmia database. The correlation coefficient between the clean signal and denoised signal can be up to 0.995, and it is greater than 0.992 even in severely contaminated situation. The sensitivity of the detecting algorithm is 99.94%, and it can be up to 99.95% in arrhythmia beats. The experiment results demonstrate that the algorithm can effectively enhance the ECG signal and detect the feature points of QRS as expected.