OBJECTIVES:  Probabilistic Sensitivity Analysis (PSA) investigates the impact of uncertainty around parameter estimates on model results. Many health economics models wrongly use the distribution of the underlying parameters to model the mean estimates in PSA, such as gamma for costs, log-normal for rates and beta for probabilities. However the appropriate distribution to sample mean values is independent of the type of parameter and is always the normal distribution given sufficient sample size. We advocate modeling guidelines should recommend the use of normal distribution when sampling means, when the sample size is large enough. We demonstrate how the PSA results change when the underlying distribution is used rather than the normal distribution. METHODS:  We used simulation methods to produce several histograms of means, each calculated from samples of increasing size, and compared the distribution of these means against the distributions that are frequently used instead of the normal distribution. We then proceeded to assess the impact of the choice of distribution on the cost-effectiveness results when the means are sampled with a normal distribution vs. the underlying distribution. RESULTS:  Histograms show that as the sample size increases, the distribution of means calculated using the underlying distribution deviates more from the distribution of the sampled means, which converge in distribution to normal. The model results suggest that using normal distribution instead of the underlying distribution changes the probability of being cost effective. CONCLUSIONS:  Sampling means from a normal distribution is a more accurate way of representing uncertainty around mean estimates. The use of the underlying distribution to sample from the means leads to inaccurate PSA results. Given the importance of PSA results in making reimbursement recommendations, researchers should be careful to use the appropriate distribution to simulate the means, which is the normal distribution.