Given a protein sequence, how can we identify whether it is an enzyme or non-enzyme? If it is, which main functional class it belongs to? What about its sub-functional class? It is important to address these problems because they are closely correlated with the biological function of an uncharacterized protein and its acting object and process. Particularly, with the avalanche of protein sequences generated in the Post Genomic Age and relatively much slower progress in determining their functions by experiments, it is highly desired to develop an automated method by which one can get a fast and accurate answer to these questions. Here, a top-down predictor, called EzyPred, is developed by fusing the results derived from the functional domain and evolution information. EzyPred is a 3-layer predictor: the 1st layer prediction engine is for identifying a query protein as enzyme or non-enzyme; the 2nd layer for the main functional class; and the 3rd layer for the sub-functional class. The overall success rates for all the three layers are higher than 90% that were obtained through rigorous cross-validation tests on the very stringent benchmark datasets in which none of the proteins has >= 40% sequence identity to any other in a same class or subclass. EzyPred is freely accessible at http://chou.med.harvard.edu/bioinf/EzyPred/, by which one can get the desired 3-level results for a query protein sequence within less than 90 s. (C) 2007 Elsevier Inc. All rights reserved.