Abstract: In this paper, various vehicles in urban streets are classified into 4 classes according to their noise levels. When traffic volume is small, traffic flows consisting,of these vehicles are regarded as a composite Poisson stream. The relationship between traffic volumes of various classes and kerbside statistical noise levels is analysed quantitatively and formulated. In consideration of traffic volumes of different kinds of vehicles, three tables for predicting kerbside levels of L₁₀, L₅₀ and L₉₀ are given. Thus, a simple and practical method for predicting noise from random traffic flow consisting of multiple vehicles in urban multiple lane street is given. A series of field measurements taken at 9 sites in Beijing has shown that prediction data are consistent with measurement data.