Most eukaryotic genes are interrupted by non-coding introns that must be accurately removed from pre-mRNAs to produce translatable mRNAs. Splicing is locally guided by short conserved sequences, but genes typically contain many potential splice sites, and the mechanisms specifying the correct ones remain poorly understood. In most organisms, introns cannot be efficiently predicted on the sole basis of sequence motifs. Hence, bioinformatics methods for gene annotation have to rely on other information (search for open reading frames, cross-species conservation of protein-coding sequences). This raises the question of how eukaryotic cells recognize intron signals within pre-mRNA sequences. I will present results of the Paramecium Genome Sequencing Consortium demonstrating that eukaryotic cells universally rely on translation to control the accuracy of intron splicing.