Summary Given the increasing number of available genomic sequences, one now faces the task of identifying their functional parts, like the protein coding regions. The gene prediction problem can be addressed in several ways. One of the most promising methods makes use of similarity information between the genomic DNA and previously annotated sequences (proteins, cDNAs and ESTs). Recently, given the huge amount of newly sequenced genomes, new similarity-based methods are being successfully applied in the task of gene prediction. The so-called comparative-based methods lie in the similarities shared by regions of two evolutionary related genomic sequences. Despite the number of different gene prediction approaches in the literature, this problem remains challenging. In this paper we present a new comparative-based approach to the gene prediction problem. It is based on a syntenic alignment of three or more genomic sequences. With syntenic alignment we mean an alignment that is constructed taking into account the fact that the involved sequences include conserved regions intervened by unconserved ones. We have implemented the proposed algorithm in a computer program and confirm the validity of the approach on a benchmark including triples of human, mouse and rat genomic sequences.