Heart failure and heart valvar diseases are chronic heart disorders which are potentially diagnosed using heart sound characteristics. Heart sound components S1 and S2 exhibit significant characteristics for valvar dysfunction while pathological S3 sound is a prominent sign for heart failure in elderly people. In this paper, a new automatic detection method of the S3 heart sound is proposed. The method is build upon wavelet transform-simplicity filter which separates S1, S2 and S3 sounds from background noise enabling heart sound segmentation even in the presence of heart murmurs or noise sources. The algorithm uses physiologically inspired criteria to assess the presence of S3 heart sound components and to perform their segmentation. Heart sound samples recorded from children as well as from elderly patients with heart failure were used to test the method. The achieved sensitivity and specificity were 90.35% and 92.35%, respectively.