Inferring CYP2D6 phenotype from genotype is increasingly challenging, considering the growing number of alleles and their range of activity. This complexity poses a challenge in translational research where genotyping is being considered as a tool to personalize drug therapy. To simplify genotype interpretation and improve phenotype prediction, we evaluated the utility of an "activity score" (AS) system. Over 25 CYP2D6 allelic variants were genotyped in 672 subjects of primarily Caucasian and African-American heritage. The ability of genotype and AS to accurately predict phenotype using the CYP2D6 probe substrate dextromethorphan was evaluated using linear regression and clustering methods. Phenotype prediction, given as a probability for each AS group, was most accurate if ethnicity was considered; among subjects with genotypes containing a CYP2D6*2 allele, CYP2D6 activity was significantly slower in African Americans compared to Caucasians. The AS tool warrants further prospective evaluation for CYP2D6 substrates and in additional ethnic populations.