Existing definitions of poor response to the antiplatelet effect of clopidogrel are empiric. Bayesian classification theory is widely used to classify subjects into non-overlapping groups based on observed responses. The purpose of this analysis is to objectively define pharmacodynamic poor responders to clopidogrel using Bayesian classification methodology. An integrated database of turbidometric platelet aggregometry data (5 and 20 microM ADP) was analyzed from 112 healthy subjects who participated in three Phase 1 clinical pharmacology studies. Change in maximum platelet aggregation (DeltaMPA) from baseline and percent inhibition of platelet aggregation (IPA) were evaluated at both 4-5 and 24 hours after a clopidogrel 300 mg loading dose (LD). Clopidogrel poor responders were predefined as having a response similar to that of drug-free baseline, derived from multiple MPA values prior to clopidogrel administration. The model identified a clopidogrel poor responder as a subject who failed to achieve and maintain an IPA >or= 25% (or DeltaMPA >or= 20%) with 5 microM ADP at 4-5 and 24 hours after dosing or who failed to achieve and maintain an IPA >or= 20% (or DeltaMPA >or= 15%) with 20 microM ADP at 4-5 and 24 hours after dosing. Application of these thresholds indicated that, depending on the concentration of ADP used, 25% to 45% of subjects were classified as clopidogrel poor responders. Using thresholds from the published literature resulted in 17% to 56% of subjects being classified as poor responders. Objective thresholds for pharmacodynamic poor responders to clopidogrel should consider the concentration of the agonist used and may help assess the consistency of pharmacodynamic response to novel ADP receptor antagonists.