OBJECTIVES: To examine whether the Markov chain/decision tree approach is better than the cost function approach in modeling the pharmaceutical costs for Attention Deficit/Hyperactivity Disorder (ADHD). METHODS: Using one half sample of a nationally representative large claim data set (Pharmetrics claim database), we applied both the Markov Chain/decision tree approach and cost function approach to model the pharmaceutical costs for attention deficit /hyperactivity disorder. Then, we examined both the validity and reliability of these two models. In terms of validity, we examined whether and how the model assumptions are satisfied under each approach. In terms of reliability, we used another half sample of the claim data set to test the two models. RESULTS: More assumptions were violated in the Markov chain/decision tree model than in cost function. One example of these violated assumptions is that Markov chain /decision tree approach assumes the independence of transition probabilities. Furthermore, because that claim data do not provide many parameters required in the Markov chain/decision tree model, the Markov chain/decision tree approach has used many arbitrary estimates, which made model very unreliable. The examples of these parameters are the rate of adequate response of each medication, the rate of titration up or down, the rate of switching between medications, the tolerable rate of each medication, and drop out rate. CONCLUSIONS: The Markov chain/decision tree approach is not necessary better than cost function approach in modeling pharmaceutical costs. When modelers do not have solid estimates of those transition probabilities and when the assumptions underlying this approach are violated, cost function approach may be better than the Markov chain/decision tree approach. The limitation of cost function approach is that it does not give clinical process information as rich as the Markov chain/decision tree approach.