Analysis of heart sounds is a popular research area for non invasive identification of several heart diseases. This paper proposes a set of 88 time-frequency features along with five different methodologies for classifying normal and abnormal heart sounds. State of the art approach was applied for segregating the fundamental heart sounds. Apart from a baseline two class classifier, separate classifiers for long and short heart sounds were also explored in order to get rid of the dependency of features on the duration of the recordings. Finally, a three class classifier was explored to deal with the noisy data present in the dataset. Both balanced and unbalanced sets were considered for crating of the training models. A comparative analysis showed that, out of all the methodologies, the three class classifier based approach produces the most optimum performance by simultaneously yielding high values of both sensitivity and specificity.