The investigation of potential adverse health effects of urban traffic-related air pollution is hampered by difficulties encountered with exposure assessment. Usually public measuring sites are few and thereby do not adequately describe spatial variation of pollutant levels over an urban area. In turn, individual monitoring of pollution exposure among study subjects is laborious and expensive. We therefore investigated whether traffic characteristics can be used to adequately predict benzene, NO2, and soot concentrations at individual addresses of study subjects in the city area of Munich, Germany. For all road segments with expected traffic volumes of at least 4000 vehicles a day (n = 1840), all vehicles were counted manually for a single weekday in 1995. The proportion of vehicles in "stop-go" mode, an estimate of traffic jam, was determined. Furthermore, annual concentrations of benzene, NO2, and soot from 18 high-concentration sites (means: 8.7, 65.8, and 12.9 micrograms/m3, respectively) and from 16 school sites with moderate concentrations (means: 2.6, 32.2, and 5.7 micrograms/m3, respectively) were measured from 1996 to 1998. Statistical analysis of the data was performed using components of two different statistical models recently used to predict air pollution levels in comparable settings. Two traffic characteristics, traffic volume and traffic jam percentage, adequately described air pollutant concentrations (R2: 0.76-0.80, P < 0.0001). This study shows that air pollutant concentrations can be accurately predicted by two traffic characteristics and that these models compare favorably with other more complex models in the literature.