Schizosaccharomyces pombe pre-mRNAs are generally multi-intronic and share certain features with premRNAs from Drosophila melanogaster, in which initial splice site pairing can occur via either exon or intron definition. Here, we present three lines of evidence suggesting that, despite these similarities, fission yeast splicing is most likely restricted to intron definition. First, mutating either or both splice sites flanking an internal exon in the S. pombe cdc2 gene produced almost exclusively intron retention, in contrast to the exon skipping observed in vertebrates. Second, we were unable to induce skipping of the internal microexon in fission yeast cgs2, whereas the default splicing pathway excludes extremely small exons in mammals. Because nearly quantitative removal of the downstream intron in cgs2 could be achieved by expanding the microexon, we propose that its retention is due to steric occlusion. Third, several cryptic 5 junctions in the second intron of fission yeast cdc2 are located within the intron, in contrast to their generally exonic locations in metazoa. The effects of expanding and contracting this intron are as predicted by intron definition; in fact, even highly deviant 5 junctions can compete effectively with the standard 5 splice site if they are closer to the 3 splicing signals. Taken together, our data suggest that pairing of splice sites in S. pombe most likely occurs exclusively across introns in a manner that favors excision of the smallest segment possible.Splice site selection has been most extensively studied in higher eukaryotes (reviewed in reference 11), where abundant evidence indicates that the unit initially recognized by the splicing machinery is the exon, as proposed by Robberson et al. nearly a decade ago (53). Particularly compelling in this regard is the observation that the most common effect of a 5Ј splice site mutation is skipping of the preceding exon rather than inclusion of the mutant intron (61; reviewed in reference 6). Moreover, in the subset of cases in which a 5Ј junction mutation causes activation of a cryptic splice site rather than exon skipping, the new exon-intron boundary is almost invariably located within the preceding exon, again supporting the view that communication occurs across the exon rather than the intron. Finally, there are significant constraints on exon length in vertebrate pre-mRNAs, consistent with the proposal that the 3Ј and 5Ј splice sites on opposite sides of the exon must be recognized concurrently. Not only are the vast majority of natural internal exons in vertebrate pre-mRNAs Ͻ300 nucleotides in length (6), but expanding an exon beyond this size causes it to be skipped (53), particularly if it is surrounded by large introns (60). In contrast to the limitations on exon length, the introns in vertebrate pre-mRNAs can be extremely large (tens of kilobases [29]).Although many questions remain to be answered, several components of the machinery responsible for exon definition have been identified. First, UV cross-linking experiments ...