OBJECTIVES: The Expected Value of Perfect Information (EVPI) is becoming an increasing valuable tool to assist healthcare decision makers with their choices of new healthcare technologies. Although the use of EVPI in healthcare decision making is on the rise the study of its properties has received little attention. This study will evaluate the properties of the EVPI in the context of a currently published model for a cholesterol lowering therapy. METHODS: The properties of the EVPI were studied using a Markov chain model that evaluated the cost-effectiveness of ezetimibe co-administration with statin therapy vs statin titration (Cook et.al 2004). Simulations of the model were run for iterations of size 1,000, 10,000 and 100,000. Baseline decisions were evaluated for blocks of 1,000 iterations and the percent of correct optimal decisions as well as the mean and standard deviation for the net benefits were tabulated within each block. RESULTS: The estimated EVPI for 1,000, 10,000 and 100,000 iterations were $19,696, $21,617and $21,648 respectively. Based on the estimated mean and standard deviations of net benefits for each treatment group we estimated that 11,325 iterations would be necessary to provide stables estimates of the EVPI for this example. Additional theoretical calculations show that the EVPI has a positive bias in small samples and is dependent on the probability of choosing the wrong treatment in any given iteration. CONCLUSION: A positive bias exists in the EVPI estimates for smaller number of iterations. This bias can be overcome by using sample size calculations to determine the appropriate number of iterations. The appropriate number of iterations will eliminate the chance of making the wrong baseline decision.