For ECG signal analysis, a QRS detection algorithm is very important. The QRS detection algorithm consists of two steps, i.e., ECG preprocessing and ECG beat detection. In preprocessing step, noises in ECG signals are removed. The higher signal to noise ratio (SNR) after noise removal in preprocessing step leads to the less complicated algorithm in beat detection step and the increase in accuracy. However, ECG signals have various types in the real situation such as normal beat and premature ventricular contraction (PVC) beat. Each type of beat has its own frequency response. Therefore, we propose the dual-band continuous wavelet transform to maximize the SNR of ECG signals after noise removal in this paper. The proposed algorithm was evaluated with the ECG signals from MIT-BIH arrhythmia database. Results demonstrate the feasibility of the method.