Recent advances in the ability to discriminate between homologous and non-homologous proteins in the "Twilight Zone" of sequence similarity, must be accompanied by accurate alignments if they are to be of value to molecular modelers. Pairwise alignments require a measure of evolutionary distance, traditionally modeled using global amino acid substitution matrices. But real differences in the likelihood of substitutions may exist for different structural contexts within proteins, since structure contributes to the selective pressure. HMMSUM (HMMSTR-based SUbstitution Matrices) is a new model for structure-dependent amino acid substitution probabilities consisting of a set of 281 matrices, one for each of the sequence-structure contexts defined in HMMSTR (a Hidden Markov Model for protein STRucture). HMMSUM does not require the structure of the protein to be known, using HMMSTR predictions instead. Alignments using the HMMSUM compare favorably BLOSUM50 alignments when validated against curated remote homolog alignments from BAliBASE.