ABSTRACT--Interest in the use of low-copy nuclear genes for phylogenetic analyses of plants has grown rapidly, because highly repetitive genes such as those commonly used are limited in number.  Furthermore, because low-copy genes are subject to different evolutionary processes than are plastid genes or highly repetitive nuclear markers, they provide a valuable source of independent phylogenetic evidence.  The gene for granule-bound starch synthase (GBSSI or waxy) exists in a single copy in nearly all plants examined so far.  Our study  of GBSSI has three parts.  1) Amino acid sequences were compared across a broad taxonomic range, including grasses, four dicotyledons, and the microbial homologues of GBSSI.  Inferred structural information was used to aid in the alignment of these very divergent sequences.  The informed alignments highlight amino acids that are conserved across all sequences, and demonstrate that structural motifs can be highly conserved in spite of marked divergence in amino acid sequence.  2) Maximum likelihood analyses were used to examine exon sequence evolution throughout grasses.  Differences in probabilities among substitution types and marked among-site rate variation contribute to the observed pattern of variation.  Of the parameters examined in our set of likelihood models, the inclusion of among-site rate variation following a gamma distribution causes the greatest improvement in likelihood score.  3)  We performed cladistic parsimony analyses of GBSSI sequences throughout grasses, within tribes, and within genera, to examine phylogenetic utility of the gene.  Introns provide useful information among very closely related species, but quickly become difficult to align among more divergent taxa.  Exons are variable enough to provide extensive resolution within the family, but with low bootstrap support.  The combined results of amino acid sequence comparisons, maximum likelihood analyses, and phylogenetic studies underscore factors that might affect phylogenetic reconstruction.  In this case, accommodation of the variable rate of evolution among sites might be the first step in maximizing the phylogenetic utility of GBSSI.