Likelihood-based codon models of sequence evolution have been the focus of much excitement and development in recent years (Anisimova and Kosiol, 2009). Most attention has centred on the detection of positive selection in datasets (Yang and Bielawski, 2000; Bielawski and Yang, 2005; Yang, 2006), but unfortunately in many cases the link to adaptive causes can be tenuous at best (Nielsen, 2009). Given the steady proliferation of codon models, what novel approaches and insights can they offer for evolutionary studies of molecular structure and function? Ancestral reconstruction methods have proven to be a powerful and innovative approach for studying adaptive evolution of protein structure and function (Thornton, 2004). Recent advances in codon models incorporating more realistic assumptions about silent substitutions (Pond and Muse, 2005; Mayrose et al., 2007; Yang and Nielsen, 2008), however, offer the opportunity to reconstruct the evolution of synonymous substitutions, a promising but largely unexplored application of these models (Du, 2010). On the other hand, codon-based clade models of evolution (Forsberg and Christiansen, 2003; Bielawski and Yang, 2004), which were originally proposed years ago, are now gaining popularity for investigating changes in evolutionary constraint, and are increasingly being used to infer functional divergence in the evolution of gene families (for eg Hernandez-Hernandez et al., 2007; Liu et al., 2010). However, these models must be used with caution, particularly with respect to the specification of the null model in likelihood ratio tests (Weadick and Chang, in press). This chapter will consist of (1) a review of codon-based ancestral reconstruction methods, followed by an example of an application of their use in inferring synonymous evolution in mammalian rhodopsins, and (2) a review of clade models of molecular evolution, followed by a description of a recently proposed clade model likelihood test of divergence and its application to teleost short-wavelength visual pigments. Ultimately, the promise of both approaches lies in the possibility of generating specific hypotheses of molecular function, which can be then be interpreted in the context of data on molecular structure and function, particularly for genes for which a variety of biochemical assays and other functional data exist.