We propose a "genome signature" for bacterial genomes based on a triplets Markov model. Without the alignment or data preprocessing required by traditional analysis methods, the model is shown to efficiently capture identifying genomic information at both species and strain levels. Based on the model, a simple assumption-free distance measure is proposed for constructing phytogeny trees. The approach avoids problems with word frequency approaches such as balancing word length and window size. The method is shown to work successfully with both bacterial whole genome data and individual eukaryotic genes. Application of the model to phylogenetic analysis is presented.