The selection of curve number to represent watersheds with similar land use and land cover is often subjective and ambiguous. Watershed with several soil groups further complicates curve number selection process while wrong curve number selection often produces unrealistic runoff estimates. The 1954 simplified Soil Conservation Services (SCS) runoff model over-predicted runoff with significant amount and further magnified runoff prediction error toward higher rainfall depths in this study. The model was statistically insignificant with the rejection of two null hypotheses and paved the way for regional model calibration study. This paper proposes a new direct curve number derivation technique from the given rainfall-runoff conditions under the guide of inferential statistics. The technique offers a swift and economical solution to improve the runoff prediction ability of the SCS runoff model with statistically significant results. A new rainfall-runoff model was developed with calibration according to the regional hydrological conditions. It out-performed the runoff prediction of the simplified SCS runoff model and the asymptotic runoff model. The derived curve number = 89 at alpha = 0.01 level. The technique can be adopted to predict flash flood and forecast urban runoff.