Computational analysis of complete genomes, followed by experimental testing of emerging hypotheses--the area of research often referred to as 'functional genomics'--aims at deciphering the wealth of information contained in genome sequences and at using it to improve our understanding of the mechanisms of cell function. This review centers on the recent progress in the genome analysis with special emphasis on the new insights in enzyme evolution. Standard methods of predicting functions for new proteins are listed and the common errors in their application are discussed. A new method of improving the functional predictions is introduced, based on a phylogenetic approach to functional prediction, as implemented in the recently constructed Clusters of Orthologous Groups (COG) database (available at http:@www.ncbi.nlm.nih.gov/COG). This approach provides a convenient way to characterize the protein families (and metabolic pathways) that are present or absent in any given organism. Comparative analysis of microbial genomes based on this approach shows that metabolic diversity generally correlates with the genome size-parasitic bacteria code for fewer enzymes and lesser number of metabolic pathways than their free-living relatives. Comparison of different genomes reveals another evolutionary trend, the non-orthologous gene displacement of some enzymes by unrelated proteins with the same cellular function. An examination of the phylogenetic distribution of such cases provides new clues to the problems of biochemical evolution, including evolution of glycolysis and the TCA cycle.