To consider the spatial variation of errors in flood forecasting, a spatial distributed runoff correction approach (SDR) based on differential response is proposed. The method establishes the differential response relationships between discharge at basin outlet and runoff of each sub-basin. The regularized least square algorithm and stepwise approximation are used to calculate the estimated runoff errors. The estimated runoff errors are allocated to correct runoff in each sub-basin. The performances of SDR, areal mean runoff correction (AMR) and autoregressive technique (AR) on updating the Xin'anjiang Model (XAJ) predictions are compared in the Dapoling and Qilijie Basins. The results show that the SDR performs better than the AMR; and with increasing lead time the proposed method exhibits more stable correction performance than the AR. The SDR approach can decompose the discharge information at basin outlet to correct runoff in each sub-basin through the flow concentration system, which can account for the spatial variability of errors and improve real-time flood forecasting.