Expert Decision Making System in medical diagnosis is becoming increasingly popular and essential. Data mining techniques provide a convenient method for mining clinical databases which are too complex and voluminous to be processed by traditional methods. This paper proposes a new technique to mine temporal rules from the clinical dataset that can be used in early prediction of the heart disease which will minimize the risk in the patients. For this purpose, a novel neuro fuzzy technique is proposed in this paper to diagnose the severity of cardio vascular disease effectively from a set of patient records. In this proposed system, a generalized database is constructed for decision making from the reduced attribute set obtained through genetic algorithm. Next, a four layer fuzzy neural network has been proposed for efficient modeling and reasoning with temporal dependencies under uncertainty. Finally, a decision support subsystem has been constructed by extracting rules from the temporal patterns retrieved from the relationships between the trained datasets. Using this subsystem a severity based diagnosis is performed, where the disease is categorized into three levels of severity (low, mild and severe). The rules generated have been used in diagnosis of disease and to find the risk factor in the patients. This system helps to increase the number of people saved from critical risks by helping medical decisions and hence improves the patient safety. Experimental results show that the prediction accuracy of the system has increased and establishes significant prior knowledge about heart disease. © EuroJournals Publishing, Inc. 2012.