Considering the poor performance of existing real-time correction methods of flood forecasting in small mountain watersheds, this study introduced the K-nearest neighbor algorithm into the real-time correction method of flood forecasting. The real-time correction model based on the K-nearest neighbor algorithm(the KNN method)was built and the Shabu Basin, in Anhui Province, was chosen as the experimental basin. Meanwhile, the real-time correction method of back-propagation neural networks(the BP method)and the traditional error autoregression method(the AR method)were also used to analyze the correction results of correction models with the evaluation indices of the flood peak relative error and the certainty coefficient. The results showed that the KNN method improved the most on the error correction of flood peak and the BP method was more accurate. The correction ability of the KNN method improved more when the historical flood data were added to the learning sample. The KNN method can effectively avoid the defect of the AR method, that the flood peak error cannot be controlled. The KNN method is well-adapted and sensitive, has high accuracy, and can be used as an effective tool to promote real-time correction in small mountain watersheds.