Continuous monitoring for runoff pollution was conducted at an urban drainage. The Sartor and Boyd model was applied to investigate SS wash-off process from impervious surface in three rainfall events. There was a good correlation between accumulated runoff and ramoved SS loads except one event. Then pollutant runoff analysis was carried out using a distributed model. Although the simulations with the calibrated model parameter values captured general trends of pollutographs, they did not work well to express the dynamic pattern. The disagreement seemed to be caused by different behavior of pollutant wash-off from roofs and roads. It was necessary to allocate different wash-off coefficient for each surface. In addition, the concept of critical runoff rate was also introduced to initiate delayed wash-off from road surface. The modified model for wash-off process explained well dynamic SS runoff pattern in several events with different rainfall characteristics. It was found that distinction of roof and road surfaces for runoff pollution was essential to explain first flush phenomena in cases of weak rainfall.