This paper presents the development of a stochastic tornado simulation model for the United States (US). The continental of the US is subjected to more than 1,000 tornadoes each year, causing significant financial losses and social disruption. Compared to hurricanes, the damage region of a tornado is relatively small and the probability of occurrence at a given location is extremely low. Therefore, it is not feasible to use solely the observed data or tracks to quantify the tornado risk for a given structure or a city that has not been affected by historical tornadoes. In this paper, a methodology for performing stochastic simulation of tornado tracks for the US is presented. The stochastic simulation framework consists of a genesis model, which utilizes the kernel density estimation to simulate the spawn locations of tornadoes. Statistical models for tornado parameters such as track length, path width and intensity, were calibrated using the tornado database maintained by the US National Oceanic and Atmospheric Administration (NOAA) Storm Prediction Center (SPC). The developed statistical models were used to simulate 1,000,000 years of tornado tracks. The simulated tornado parameters include the tornado occurrence rate, intensity (EF-scale), location, touchdown time, path length and path width. All these parameters are geographic dependent, meaning the parameters vary depending on the tornado spawn locations. The simulated spawn rates and other key parameters for the continental of the US are compared to the observations. Good agreements are observed between simulations and observations. To illustrate a potential use of the simulated tornado track database, a probabilistic tornado hazard analysis was performed for Moore, Oklahoma. The 50-year tornado hazard curves for three domain sizes are developed to assess the influence of the domain size on tornado risk.