The gene set for Medaka was built using a modified version of the standard Ensembl genebuild pipeline. The species-specific sequence resources (medaka cDNA and protein) are very limited, so the majority of gene models are based on genewise alignments of proteins from other species that are genetically distant to medaka. To improve accuracy, genewise alignments were made to stretches of genomic sequence rather than to 'miniseqs'. Medaka cDNAs were aligned and used to add UTRs to the genewise predictions where possible. The gene models were then assessed by generating sets of potential orthologs to genes from other species. Potentially missing predictions and partial gene predictions were identified by examining the orthologs and used to improve the gene models. Finally, additional genes were predicted using medaka EST evidence, and medaka ESTs were used to add UTRs to the gene set where possible.