OBJECTIVES To estimate the uncertainty regarding adjuvant treatment selection for postmenopausal women with early stage oestrogen-receptor positive breast cancer when pre-treatment CYP2D6 pharmacogenetic testing is considered as an option. In addition, the expected value of partial perfect information (EVPPI) was estimated for different parameter sets to inform research prioritisation. METHODS A decision analytic model estimated lifetime costs and quality adjusted life years (QALYs) for four comparators: 5 years of tamoxifen; 5 years of an aromatase inhibitor (AI); CYP2D6 test and treat wt/wt genotypes with tamoxifen and other genotypes with an AI; CYP2D6 test and treat wt/wt and wt/*4 genotypes with tamoxifen and *4/*4 genotypes with an AI. No trial data for CYP2D6 contingent treatment pathways were identified. Trial data comparing tamoxifen to anastrozole was therefore synthesised with observational data linking CYP2D6 genotype to recurrence in patients receiving tamoxifen. Estimates of the EVPPI were derived by attaching distributions to input parameters and using two-level Monte Carlo simulation. EVPPI estimates were generated for parameters describing the efficacy of tamoxifen and anastrozole; parameters describing genotype-specific tamoxifen efficacy; genotype prevalence; utility weights and health state costs. RESULTS The strategy of CYP2D6 test and treat wt/wt patients with tamoxifen and all others with an AI maximised expected net benefit assuming a decision threshold of £30,000/QALY, and had an incremental cost-effectiveness ratio of £14,133/QALY. However, this maximised net benefit with only 61% certainty. This substantial decision uncertainty led to an expected value of perfect information estimate of £84 million. The EVPPI estimate for parameters describing genotype-specific tamoxifen efficacy was £57 million. Estimates for other parameter groups were low. CONCLUSIONS Further CYP2D6 genotyping studies amongst patients receiving tamoxifen should be prioritised. Expected value of sample information analysis could be used to establish the cost-effectiveness and optimal design of this primary research.