We take a preliminary look at computational methods of identifying similarity in musical compositions. We have chosen the “Raga” paradigm of Indian classical music as the basis of our formal model since it is well understood and semi-formal in nature. We address the issues of why we need a computational basis of the judgment whether two compositions are similar or not, and why traditional distance metrics fail. The Raga paradigm is discussed and the computational attributes of a raga are defined. Machine learning techniques for assimilating the tuple that formally defines a raga are explored, and matching algorithms for identifying genres of music that have been learnt are proposed. This is essentially a positional paper and we have just started testing some of our models, but current research shows a lot of promise.