The promise of plentiful data on common human genetic variations has given hope that we will be able to uncover genetic factors behind common diseases that have proven difficult to locate by prior methods. Much recent interest in this problem has focused on using haplotypes (contiguous regions of correlated genetic variations), instead of the isolated variations, in order to reduce the size of the statistical analysis problem. In order to most effectively use such variation data, we will need a better understanding of haplotype structure, including both the general principles underlying haplotype structure in the human population and the specific structures found in particular genetic regions or sub-populations. This paper presents a probabilistic model for analyzing haplotype structure in a population using conserved motifs found in statistically significant sub-populations. It describes the model and computational methods for deriving the predicted motif set and haplotype structure for a population. It further presents results on simulated data, in order to validate the method, and on two real datasets from the literature, in order to illustrate its practical application.