P wave detection requires a robust QRS-T cancellation method. Interesting algorithms have been proposed for beat-to-beat QRS-T cancellation. Previous studies have shown that adaptive methods lead to good cancellation of the QRS-T interval which generally guarantees the performance of P wave detection. However, adaptive methods suffer from the nonstationary behavior of ECG signals and particularly beat-to-beat morphology changes of the QRS. We present a new approach for two ECG channel QRS-T cancellation based on the dyadic wavelet transform. The method is insensitive to QRS morphology changes and performs well in the presence of ectopic beats, transient artifacts, baseline drifts and isolated P waves. Our approach allows the P wave to be enhanced better than methods recently proposed.