Studies of expressed sequence tag data sets have revealed large numbers of splicing variants for human genes, but it remains challenging to distinguish functionally important variants from aberrant splicing, clarify the nature of the alternative functions, and understand the signals that regulate splicing choices. To help address these issues, we have constructed and analyzed a large data set of 1,478 exon-skipping alternative splicing (AS) variants evolutionarily conserved in human and mouse. In about one-fifth of cases, one isoform appears subject to nonsense-mediated mRNA decay (NMD), supporting the idea that a major role of AS is to regulate gene expression; one-quarter of these NMD-inducing cases involve a conserved exon whose apparent sole purpose is to mediate destruction of the message when included. We explore sequence conservation likely related to splicing regulation, using in part a measure of the overall amount of conserved information in a sequence, and find that the increased conservation that has been observed within AS exons primarily affects synonymous sites, suggesting that regulatory signals significantly constrain synonymous substitution rates. We show that a lower frequency of the inclusion isoform relative to the exclusion isoform tends to be associated with weaker splice site signals, smaller exon size, and higher intronic sequence conservation, and provide evidence that all of these factors are under selection to control relative isoform frequencies. Some conserved instances of AS appear to represent aberrant splicing events that by chance have occurred in both species, and we develop a nonparametric likelihood approach to identify these.T wo known roles for alternative splicing (AS) are to allow several proteins to be produced from a single gene, and to reduce gene expression by yielding isoforms that are degraded by nonsensemediated mRNA decay (NMD) or other mechanisms (1, 2). Genome-wide studies using cDNA and EST databases have suggested that 22-74% of mammalian genes are alternatively spliced, with perhaps 35% of AS isoforms predicted to be subject to NMD (2-8). However, the biological significance of these estimates remains unclear, because many database variants may represent aberrant splicing events lacking a functional role (ref. 9 and references therein). Given that the primary function of NMD is presumably to remove aberrant transcripts, it is particularly difficult to assess the significance of NMD isoforms. Filtering out variants that are supported by few ESTs (5, 6), that have been found only in tumor cells or cell lines (10), or other anomalies (9), is not guaranteed to catch all aberrant splicing cases, and may incorrectly eliminate functional rare variants. Determining which variants are biologically important consequently remains a major challenge.An important approach to identifying functionally important AS is to focus on isoforms common to orthologous genes in two species (11)(12)(13)(14). This method does not eliminate the possibility of false positives representing...