Heart abnormalities can be found from the heart sound. This paper describes a new approach for classifying phonocardiograms. The approach relies on type-3 fuzzy logic (T3FL) and fractal dimension (FD). Previous works have shown that FD produces different values for different types of heart diseases. Based on these facts, a T3FL is used to contain expert knowledge in heart disease classification by using the FD values. In this case, T3FL enables uncertainty representation of heart sound classification. Results show the proposed approach effectiveness.