Identifying the polymorphisms that contribute to disease predisposition and drug response is a major goal of the post-genome era. Single nucleotide polymorphisms (SNPs) in disease-related genes are often used as candidates in the search for causative variations. Association tests based on haplotypes have also been suggested and, at times, have provided greater statistical power than tests based on the underlying SNPs. Here we review the statistical model traditionally used to describe association studies for complex traits and derive novel results for the relative power of SNP-based and haplotype-based tests of association. In the model, a set of independent SNP-based variations, some of which contribute to a measured phenotype, may be used as markers directly or may be organised into haplotype markers. Provided that the marker set includes all the causative SNPs, we find a simple rule for the relative power of SNP and haplotype markers: SNP-based tests have greater power when the number of causative SNPs (a subset of the total set of SNPs) is smaller than the total number of haplotypes. Furthermore, we find that regression tests for the simple main effect of each haplotype are generally more powerful than ANOVA tests applied to haplotype pairs. A review of recent literature supports our findings.