Numerous genome sequence projects of various organisms have resulted in generation of large amount of data on genes and proteins sequence information. Functional annotation of these proteins is important to bridge gap between sequence information and functional characterization. As experimental approaches for characterizing the functional class of an enzyme are expensive and time consuming, computational prediction methods are an effective alternative. Various approaches like homology-based function transfer and machine learning methods have been utilized for in silico enzyme functional classifications in terms of Enzyme Commission number (EC number) for a protein. Different types of features have been used in various machine learning techniques and each has its own advantages and limitations. The critical evaluation of performance measure in terms of predictive ability of these methods is necessary. Here, a systematic review on the various approaches used by different research groups, their utility and inference is presented.