Background: ECG signal is a weak bio-electricity signal, and can be influenced by baseline wander, power line interference, electromyographic (EMG) signal and so on. The baseline wander is a low-frequency interference and there are a series of well-deleloped methods to remove it. The power interference has fixed frequency and can be eliminated by different types of trap filter. However, EMG noise has a broad bandwidth overlapping on the ECG signal, which is hard to suppress. This research uses one-dimensional Kalman filter to remove EMG noise after preliminary filtering and QRS complex wave recognition of real-time ECG signal. Methods: Kalman filter is recursive and can run in real time. While in prediction step, Kalman filter produces estimates of the current variables, along with their uncertainties. Once the outcome of next measurement is observed, these estimates are updated. In this research, the low pass and notch FIR filter are used firstly to suppress power line and high frequency interference. Then a median filter is used to delete baseline wander. A Kaiser window is also used to prevent spectrum leakage. After these preprocessing, the wavelet transform method is used to initially identify the R peaks, as well as Q peaks and S peaks. Since EMG noise is similar as white noise as to ECG, the estimate of EMG noise is simple and Kalman filter is suitable to remove EMG in real time. Results: We test the 48 ECG data in MIT arrhythmia database and 1475 ECG data collected by a portable ECG card. Without weakening the R peaks, the EMG noise is suppressed successfully, while T-wave and P-wave can be automatically identified with the smooth signal. The sensitivity of QRS recognition is above 99%. What’s more, the sensitivity for premature ventricular contraction automatic identification of MIT arrhythmia database is also above 99%.