Gene prediction is an essential step in understanding the genome of a species once it has been sequenced. For that, a promising direction in current research on gene finding is a comparative genomics approach. In this paper, we present a novel approach to identifying evolutionarily conserved protein-coding sequences in genomes. The method takes advantage of the specific substitution pattern of coding sequences together with the consistency of reading frames. It has been implemented in a software called PROTEA. Large-scale experimentation shows good results. PROTEA is intended to be a useful complement to existing tools based on homology search or statistical properties of the sequences.