The parameterizations of atmospheric turbulence dispersion in many commonly used numerical models are based on the similarity relations in highly flat and sparsely populated areas. Such parameterizations limit the applicability of the models for simulating heavy haze processes that frequently occur in densely populated urban areas with complex terrains. In this study, we examined the similarity relations of key turbulence dispersion parameters in the Lagrangian dispersion models, including the turbulent standard deviations, high-order moments (the third and fourth moments), and integral time scales for longitudinal, lateral, and vertical wind velocity components during heavy haze days in downtown Beijing, China. The similarity relations were analyzed in the framework of the local similarity theory, which is commonly used in studies of urban turbulence. It was found that in highly stable or unstable conditions, the normalized turbulent standard deviations and high-order moments (the normalized integral time scales) during heavy haze days generally increased (decreased) significantly slower than those during clean days. However, in nearly neutral conditions, the normalized values for heavy haze and clean days were approximately equal. The results indicate that the development of parameterizations of turbulence dispersion during heavy haze days, especially in highly stable or unstable conditions, warrants greater research focus.