Pairwise sequence alignment forms the basis of numerous other applications in bioinformatics. The quality of an alignment is gauged by statistical significance rather than by alignment score alone. Therefore, accurate estimation of statistical significance of a pairwise alignment is an important problem in sequence comparison. Recently, it was shown that pairwise statistical significance does better in practice than database statistical significance, and also provides quicker individual pairwise estimates of statistical significance without having to perform time-consuming database search. Under an evolutionary model, a substitution matrix can be derived using a rate matrix and a fixed distance. Although the commonly used substitution matrices like BLOSUM62, etc. were not originally derived from a rate matrix under an evolutionary model, the corresponding rate matrices can be back calculated. Many researchers have derived different rate matrices using different methods and data. In this paper, we show that pairwise statistical significance using rate matrices with sequence-pair-specific distance performs significantly better compared to using a fixed distance. Pairwise statistical significance using sequence-pair-specific distanced substitution matrices also outperforms database statistical significance reported by BLAST.