Cardiac arrhythmia detection and recognition have been major topics for electrocardiography. In this research, we develop a recognition system for a variety of arrhythmias. Alpha-stable distribution, with a general form of Gaussian distribution, shares more properties with arrhythmia such as sharp spikes or occasional impulses of outlying observations than one would expect from normally distributed signals. Since some electrocardiograms (ECGs) with arrhythmias are chaotic and difficult to interpret in the time domain, demonstrating biomedical data with statistical information would provide an alternative for exploring more underlying characteristics. In addressing this problem, a novel method of accurate parameter estimation of cardiac signal is proposed. In addition, based on various arrhythmia types, a classifier using the support vector machine (SVM) is designed to verify the usefulness of our methods. The developed system is shown to yield more accurate modeling, and therefore a higher identification rate compared to the Gaussian-based approach.