OBJECTIVES: Current procedures to assess costs of drugs for use in comparative effectiveness studies fail to account for important issues related to pharmacy practice such as availability of generic drugs, differing manufacturer and package sizes, and clinically equivalent and reasonably interchangeable dosages.  This work proposes a method aimed at better estimation of drug costs that includes a level of uncertainty.   METHODS: Using patient drug histories and the Micromedex Redbook costs database, we construct an algorithm which accurately matches every prescription drug order to the full set of therapeutically appropriate National Drug Codes (NDCs) that could be used to fill each prescription.  The algorithm, calculated on a per unit of ingredient, is flexible and can be modified to include more or less inclusive sets of possible NDC codes based on various definitions of reasonably interchangeable drug formulations. We compare a (i) simple method, using only a single NDC code, (ii) complete method that matches every potentially suitable NDC code, and (iii) pill rationing to favor exact doses.  The complete and pill rationing methods introduce uncertainty between what was prescribed and what NDC code could be used to fill each prescription. RESULTS:  As a concrete example using 500mg dose of Cephalexin, our simple method yields an estimated cost of $2.74 per dose, our complete method with 556 NDCs yielded mean (sd) $3.61 (2.40), and  pill rationing  with 314 NDCs yielded mean (sd) $3.55 (1.73). CONCLUSIONS: This method highlights several important issues when assessing costs of a drug included in comparative effectiveness studies.  It also proposes a way to better account for uncertainty when modeling estimated costs and more consistency in including individuals using the same drug dosage at the same cost.