Protein function prediction is important for understanding life at the molecular level and therefore is highly demanded by biomedical research and pharmaceutical applications. To overcome the problem, in this work I have proposed a novel multi-label protein function prediction based on hierarchical approach. In Hierarchical multi-label classification problems, each instance can be classified into two or more classes simultaneously, differently from conventional classification. Mainly, the proposed methodology is consisting of three phases such as (i) Creating clusters (ii) Generation of class vectors and (iii) classification of instances. At first, we create the clusters based on the hybridization of KNearest Neighbor and Expectation Maximization algorithm (KNN+EM). Based on the clusters we generate the class vectors. Finally, protein function prediction is carried out in the classification stage. The performance of the proposed method is extensively tested upon five types of protein datasets, and it is compared to those of the two methods in terms of accuracy. Experimental results show that our proposed multi-label protein function prediction significantly superior to the existing methods over different datasets.