OBJECTIVE: The diversity of procedure in pharmaceutical research requires a guideline to choose appropriate matching method. Coherent guidelines for practice are absent. In this paper we evaluate the several matching techniques and provide a guideline to choose the best. METHODS: We proposed the following ways to check for the balance: 1) Two sample t-statistic between the mean of the treatment group for each explanatory variables with the mean of these variables in the control group; 2) The mean difference as a percentage of the average standard deviations; 3) Percent reduction bias in means of explanatory variables after matching and initially; 4) Compare treatment and control density estimates for the explanatory variables; and 5) Compare the density estimates of the propensity scores of control units with that of the treated units. RESULTS: Medstat Market Scan data used to provide empirical examples. We examined 2 to 1 matching, nearest neighborhood matching (NNM) with replacement, NNM without replacement, MM matching (MM), MM with calibers, stratification method, kernel matching and radius matching. Comparing techniques according to the above criteria yield that 2 to 1 and NNM without replacement provides the worst balance. The difference between the control and treatment variables was significant. To choose among the rest, we estimated the average treatment effect according to each matching procedures and calculated the deviation from the mean of estimated average treatment effect. MM with calibers where calibers is selected as a quarter of standard deviation of estimated propensity score provides least deviation, there this procedure was superior to the others. CONCLUSION. Sensitivity analysis of the matching techniques is especially important since none of the proposed methods in literature is a priori superior to the others. The joint consideration offers a way to assess the robustness of the estimates.