The automatic classification of proteins into groups is one of the major objectives for mining the increasing amount of data released by genomic and metagenomic sequencing projects. Ortholog and in-paralog accurate classification is motivated by the notion of descriptive biology. Facing the tremendous quantity of very complex protein datasets, one way to understand biological function, structure conservation as well as evolution history is to associate or group them into classes according to their sequence homology, function, folding motifs and structural features. In this review, we will explore and compare the different approaches and databases of automatic clustering and classification developed in the last years. We will also discuss the impact of hierarchies and clusters of proteins to protein function and phylogeny predictions.