The identification of pseudogenes is an integral and significant part of the genome annotation because of their abundance and their impact on the experimental analysis of functional genes. Most of the computational annotation systems are not optimized for systematic pseudogene recognition, often annotating pseudogenes as functional genes, and users then propagate these errors to subsequent analyses and interpretations. In order to validate gene annotations and to identify pseudogenes that are potentially mis-annotated, we developed a novel approach based on whole genome profiling of existing transcript and protein sequences. This method has two important features: (i) equally detects both processed and non-processed pseudogenes and (ii) can identify transcribed pseudogenes. Applying this method to the human Ensembl gene predictions, we discovered that 2011 (9% of total) Ensembl genes in the categories of known and novel might be pseudogenes based on expression evidence. Of these, 1200 genes are found to have no existing evidence of transcription, and 811 genes are found with transcription evidence but contain significant translation disruption. Approximately 40% of the 2011 identified pseudogenes presented a multi-exon structure, representing non-processed pseudogenes. We have demonstrated the power of whole genome profiling of expression sequences to improve the accuracy of gene annotations.