In many Asian countries, physicians both prescribe and dispense drugs. This practice is hypothesized to have caused high drug expenditure and widespread prescription of antibiotics in Asia. Recently, Taiwan implemented the separation policy on an experimental basis. This paper's objective is to empirically evaluate the impact of Taiwan's reform to separate drug prescribing and dispensing on drug expenditure and total health expenditure. The research design consists of a pre/post comparison of the experimental with the control sites (difference-in-difference). Separation policy was implemented in Kaohsung and Taipei in March 1997, and expanded to Chia-yi and Taichung in March 1998. Changes in drug prescription behaviour before and after implementation in these two pairs of experimental cities were compared to Hsin-chu and Tainan (control), where separation policy was not implemented during the study period. To reduce resistance, providers in experimental sites were allowed to hire on-site pharmacists and dispense drugs through them if they chose to do so. Our study sample consists of all outpatient visits to clinics in the study sites between December 1996 and June 1998, with a total of 55.23 million claim records. The drug prescription rate, drug expenditure and total health expenditure per visit were the main outcome measures. We found that the probability of prescription and drug expenditure per visit were, respectively, 17-34% and 12-36% less among visits to clinics without on-site pharmacists, compared with the control sites. However, no difference in total health expenditure was found between these two types of visits. Hence, the separation policy could be effective in reducing drug expenditure and affecting prescription behaviour, but is less certain as a policy for reducing total health expenditure. We also found that the policy has practically no effect on clinics that have on-site pharmacists.