Decision-making in traffic regulations is challenging with uncertainty in the environment. In this research we extend probabilistic engineering design concepts to policy decision-making for urban traffic with variability from field data. City traffic is simulated using user equilibrium and cellular automata. A cellular automata (CA) model is developed by combing existing CA models with tailored rules for local traffic behaviors in Tainan, Taiwan. Both passenger sedans and motorcycles are considered with the possibility of passing between different types of vehicles. The tailpipe emissions from all mobile sources are modeled as Gaussian dispersion with finite line sources. Speed limits of all roads are selected as independent policy design variables, resulting in a problem with 50 dimensions. We first study the impacts of a particular policy-setting on traffic behaviors and on the environment under various sources of uncertainties. The genetic algorithm, combined with probabilistic analysis, is then used to obtain the optimal regulations with the minimal cost to the environment in compliance to the current ambient air quality standards.