We have developed a method for prediction of eukaryotic gene structures (exon-intron organizations) based on sequence homology together with several lines of statistical information such as coding potential and signal strength around exon-intron boundaries and translational start sites. Based on testing upon sets of human and C. elegans genes of known structures, we found that the method predicts coding nucleotides at an accuracy of 96% or more, when the amino acid identity between the reference and target sequences exceeds 50%. Using this method, we are attempting to identify all genes involved with drug metabolism in organisms whose entire genomic sequences have been determined.