Estimates of the number of genes of organisms with a wide range of biological complexity will emerge from genome sequencing projects, and this information will be crucial in correlating gene number with biological complexity. Computer-based annotations of the genome sequences of Saccharomyces cerevisiae, Caenorhabditis elegans and Drosophila melanogaster predict approximately 6,000, 19,000 and 13,600 genes, respectively. It is paradoxical that C. elegans contains 50% more genes than D. melanogaster and that the latter contains only twice as many genes as S. cerevisiae. However, these numbers should be considered with caution until gene predictions are verified using experimental approaches. To verify gene predictions experimentally, we amplified predicted open reading frames from a C. elegans complementary DNA library using the polymerase chain reaction, cloned them and generated an open reading frame sequence tag (OST) for each of them. In a random sample (n=1,200) of the approximately 10,00 genes predicted from the entire C. elegans genome but experimentally unverified so far, 90% of these predicted genes could be unambiguously verified by OST sequencing. When added to the list of approximately 9,500 genes previously verified by individual laboratories, by expressed sequence tag sequencing or both, our data confirm the existence of at least 19,000 protein-encoding genes in C. elegans. We suggest that ORFeome cloning coupled with OST analysis could be used to verify gene number predictions for other organisms. Furthermore, the demonstration and identification of 10,000 new genes provides a resource for comparative genomics and gene expression studies. We also discuss our findings in the context of the current debate on the human gene number.