The first version of the Caenorhabditis elegans ORFeome cloning project, based on release WS9 of Wormbase (August 1999), provided experimental verifications for approximately 55% of predicted protein-encoding open reading frames (ORFs). The remaining 45% of predicted ORFs could not be cloned, possibly as a result of mispredicted gene boundaries. Since the release of WS9, gene predictions have improved continuously. To test the accuracy of evolving predictions, we attempted to PCR-amplify from a highly representative worm cDNA library and Gateway-clone approximately 4200 ORFs missed earlier and for which new predictions are available in WS100 (May 2003). In this set we successfully cloned 63% of ORFs with supporting experimental data ("touched" ORFs), and 42% of ORFs with no supporting experimental evidence ("untouched" ORFs). Approximately 2000 full-length ORFs were cloned in-frame, 13% of which were corrected in their exon/intron structure relative to WS100 predictions. In total, approximately 12,500 C. elegans ORFs are now available as Gateway Entry clones for various reverse proteomics (ORFeome v3.1). This work illustrates why the cloning of a complete C. elegans ORFeome, and likely the ORFeomes of other multicellular organisms, needs to be an iterative process that requires multiple rounds of experimental validation together with gradually improving gene predictions.