Sean Eddy wrote an excellent explanation of the BLOSUM Matrix in his Nature Biotechnology paper Where did the BLOSUM62 alignment score matrix come from? He also gives an real examples of how one would calculate the score for different amino acid pairs and where the numbers come from. Further references given at the end of the paper might also be useful reading, specifically: Amino acid substitution matrices from an information theoretic perspective, Methods for assessing the statistical significance of molecular sequence features by using general scoring schemes, and Amino acid substitution matrices from protein blocks.