We present here a novel methodology for the identification of genome regions potentially spanning one or more protein coding genes. It is based on the detection of clusters of conserved sequence tags whose evolutionary dynamics, based on the observation of an excess bias of synonymous substitutions at nucleotide level and of conservative replacements at protein level, suggests a likely protein coding role. A benchmark test carried out on a 236 Mbp of human-mouse syntenic regions from human chromosomes 15, 21 and 22 identified 25 CST clusters potentially containing unannotated genes. A further annotation update of the human genome assembly revealed that 11/25 clusters actually contained a total of 20 validated genes and 10 of the remaining 14 clusters had several experimental evidence in support of the presence of protein coding genes. These findings demonstrate the effectiveness and high prediction reliability of the proposed methodology which could specifically be applied to the annotation of novel genome sequences.