In this study, an algorithm for arrhythmia classification that conforms to the recommended practice by the Association for the Advancement of Medical Instrumentation (AAMI) is presented. Our approach efficiently exploits the inherent sparse representation of the electrocardiogram (ECG) signals. It involves, first, designing a separate dictionary for each Arrhythmia class. To this end, the alternating direction method of multipliers (ADMM) and the K-SVD, a dictionary learning algorithm based on singular value decomposition (SVD) approach, are applied. Sparse representations, based on new designed dictionaries, of each new test QRS complex are then calculated and assigned to the class associated with the largest Pietra Index (PI) afterwards. Our experiments showed promising results with accuracies ranging between 80 % and 100 %.