OBJECTIVES: In economics the triangular distribution is often used when limited information is available for different parameters. This is also the case in health economic modelling when performing probabilistic sensitivity analysis (PSA). In PSA, distributions are assigned to input parameters in order to assess the uncertainty in the model. One of the main criticisms of PSA is that the distributions can be chosen arbitrarily. Analysts can thereby manipulate the choice of distributions to bias the results.  This study investigates the usage of the triangular distribution for describing the uncertainty of relative risks (RR) compared to the lognormal distribution and empirical distribution RR generated through simulations. METHODS: Ten thousand simulations of the triangular distribution, the log normal distribution and relative risks constructed from two binomial distributions were performed. Descriptive statistics and graphical plots were constructed. RESULTS: The triangular distribution does not have the support of the full positive real axis and as such extreme values, such as very small and very large numbers, have a zero probability of being measured. However, values around the mode are prone to be drawn with a higher probability compared to both the exact values and the log normal distribution. The lognormal distribution tends to overestimate the RR compared to the empirical distribution. CONCLUSIONS: This study shows that the triangular distribution is a poor choice for characterizing the uncertainty of RR. The overestimation of the RR can introduce bias, for instance, if used for responder rates or death rates. The lognormal distribution appears to be a better approximation, but if the actual number of events and total number of exposed are available, the empirical simulation is of course preferred.