The Curve Number (CN) method is widely used for predicting surface runoff during storm events. However, there are still uncertainties in the method, particularly regarding of an initial abstraction (λ) standard value of 0.2 and on the selection of the most suitable CN values. This paper computes the CN values for 57 Brazilian watersheds with areas below 250 km2, using a 30-year historical daily time series of rainfall and runoff data from the Catchments Attributes for Brazil (CABra). CN values were determined using five methods: median, arithmetic mean, geometric mean, least squares (non-linear fit) and the Asymptotic method. Furthermore, the study investigates the performance of estimating surface runoff by using two initial abstraction coefficient values (λ = 0.2 and 0.05). The results indicate that least squares fit method with λ 0.05 showed the best runoff performance estimation, with Nash-Sutcliffe Efficiency (NSE) values greater than 0.5 for 42% of the studied catchments. Tabulated CN values did not show satisfactory results (NSE < 0.5 for 98% of the studied catchments), which indicates the need of a local calibration of the CN. The results suggest the improvements in runoff estimation by using λ = 0.05 instead 0.2, with a S0.2 to S0.05 conversion factor of 2.287. In 32% of the studied catchments was observed Complacent behaviour using λ 0.2. This occurred in catchments where runoff mechanisms are dominated by subsurface stormflow. In general, the SCS-CN method provided suitable results of runoff estimation, being its performance better in catchments that presented greater runoff generation.
