Some genes that contain premature nonsense codons express alternatively-spliced mRNA that has skipped the exon containing the nonsense codon. This paradoxical association of translation signals (nonsense codons) and RNA splicing has inspired numerous explanations. The first is based on the fact that premature nonsense codons often reduce mRNA abundance. The reduction in abundance of full-length mRNA then allows more efficient amplification during PCR of normal, minor, exon-deleted products. This mechanism has been demonstrated to explain an extensive correlation between nonsense codons and exon-skipping for the hamster Hprt gene. The second explanation is that the mutation producing an in-frame nonsense codon has an effect on exon definition. This has been demonstrated for the Mup and hamster Hprt gene by virtue of the fact that missense mutations at the same sites also are associated with the same exon-deleted mRNA. The third general explanation is that a hypothetical process takes place in the nucleus that recognizes nonsense codons, termed 'nuclear scanning', which then has an effect on mRNA splicing. Definitive evidence for nuclear scanning is lacking. My analysis of both nonsense and missense mutations associated with exon skipping in a large number of genes revealed that both types of mutations frequently introduce a T into a purine-rich DNA sequence and are often within 30 base pairs of the nearest exon boundary. This is intriguing given that purine-rich splicing enhancers are known to be inhibited by the introduction of a T. Almost all mutations associated with exon skipping occur in purine-rich or A/C-rich sequences, also characteristics of splicing enhancers. I conclude that most cases of exon skipping associated with premature termination codons may be adequately explained either by a structural effect on exon definition or by nonquantitative methods to measure mRNA, rather than an effect on a putative nuclear scanning mechanism.