A risk analysis tool called Probabilistic Fire Simulator (PFS) is developed for the computation of the distributions of fire model output variables and the sensitivities of the output variables to the inputs. PFS performs a Monte Carlo simulation using different fire models, including CFAST two-zone model and FDS fluid dynamics model. In this work, a new technique is developed for the use of two different fire models in the same Monte Carlo simulation. The two-model Monte Carlo technique provides a computationally effective means to improve the accuracy of the fast but inaccurate models, using the results of the more accurate but computationally more demanding models. The technique is tested in three scenarios: approximation of an analytical function, calculation of a ceiling jet temperature and a simulation of a simple room fire.