Abstract: This study presents a multi-stage system for reliable heart rhythm monitoring and diagnosis. It is comprised of three components including data pre-processing and feature extraction, abnormal arrhythmia detection and diagnosis. In the first stage, three different feature extraction methods are applied together to obtain a composite representation of the ECG waveform. In the second stage, the Multivariate Statistical Process Monitoring (MSPM) approach is used to capture the natural variations of the normal cardiac state and to detect any abnormal arrhythmia. Then a feed-forward neural network is used to classify the abnormal arrhythmia in 5 different classes. The results of experiments show the good performance of the proposed system.