Many human diseases, including Fanconi anemia, hemophilia B, neurofibromatosis, and phenylketonuria, can be caused by 5Ј-splice-site (5Јss) mutations that are not predicted to disrupt splicing, according to position weight matrices. By using comparative genomics, we identify pairwise dependencies between 5Јss nucleotides as a conserved feature of the entire set of 5Јss. These dependencies are also conserved in human-mouse pairs of orthologous 5Јss. Many disease-associated 5Јss mutations disrupt these dependencies, as can some human SNPs that appear to alter splicing. The consistency of the evidence signifies the relevance of this approach and suggests that 5Јss SNPs play a role in complex diseases.[Supplemental material is available online at www.genome.org.]The sequenced genomes of a wide range of organisms allow global, comparative analyses of regulatory sequences. The genomic set of splice-site sequences corresponds to a large-scale splicing experiment performed by nature under evolutionary constraints. Here we focus on 5Ј-splice-site (5Јss) sequences of the U2-type GT-AG class, which comprise over 98% of all splice sites, and use disease-causing mutations, human single nucleotide polymorphisms (SNPs), and variations in natural splice sites in the genome (within and between species) to infer properties inherent to 5Јss, with important implications for human genetics.Splice sites are conserved sequences at both ends of an intron that are recognized during the initial steps of splicing (Hastings and Krainer 2001;Brow 2002;Jurica and Moore 2003). Both the 5Јss and the 3Ј splice site (3Јss) conform to degenerate motifs that are recognized by specific splicing factors. The U2-type GT-AG 5Јss, spanning 3 nucleotides (nt) at the 3Ј end of the exon and 6 nt at the 5Ј end of the intron, is initially recognized via base pairing to the 5Ј end of the U1 snRNA (Fig.