Recent comparative genomic analysis of alternative splicing has shown that protein modularity is an important criterion for functional alternative splicing events. Exons that are alternatively spliced in multiple organisms are much more likely to be an exact multiple of 3 nt in length, representing a class of "modular" exons that can be inserted or removed from the transcripts without affecting the rest of the protein. To understand the precise roles of these modular exons, in this paper we have analyzed microarray data for 3,126 alternatively spliced exons across ten mouse tissues generated by Pan and coworkers. We show that modular exons are strongly associated with tissue-specific regulation of alternative splicing. Exons that are alternatively spliced at uniformly high transcript inclusion levels or uniformly low levels show no preference for protein modularity. In contrast, alternatively spliced exons with dramatic changes of inclusion levels across mouse tissues (referred to as "tissue-switched" exons) are both strikingly biased to be modular and are strongly conserved between human and mouse. The analysis of different subsets of tissue-switched exons shows that the increased protein modularity cannot be explained by the overall exon inclusion level, but is specifically associated with tissue-switched alternative splicing.