An ectopic heartbeat initiated by the ventricles is considered as Premature Ventricular Contraction (PVC) beat. For any individual, unifocal PVCs are typically monomorphic, whereas multifocal PVCs have polymorph contour. The ectopic rate especially of multimorphic PVCs is significantly associated with sudden death and many other main arrhythmic events. In order to group PVCs upon their morphology, a robust clustering method has been developed. In this work, the already existing approach of combining Principal Component Analysis (PCA) and Self Organizing Map (SOM) for patient specific beat clustering is used and optimized to deal with a variable number of leads and to cluster PVC beats in a noisy environment. The algorithm is tested on manually annotated multi-lead records using three leads.