Papers and reports on investigations of the urban drainage in Japan were collected (1968–2011) and a database of urban runoff loading for each measured precipitation event was prepared. The targeted pollutants were BOD, COD, SS, TN, and TP, and their event mean concentrations (EMC) were included in each data set along with their precipitation event conditions. The distribution of loading and the geographic conditions of the basins were clarified, and the dependence of the EMC values on the precipitation and geographic conditions was modeled using the regression equation of a power law. Using this relation, the predicted yearly loading and coefficients of variation were derived for four representative precipitation conditions in Japan. The required sampling frequency to attain a certain accuracy of the annual loading was estimated by a simulation using the above model. From the results, at least half of the precipitation events must be sampled for the 95% confidence interval of annual loading to be within the range of half to double the true value.