The prediction skills for climate models can be improved by using the statistical downscaling method. In order to gain better and objective midsummer precipitation prediction in Shanxi, a statistical downscaling method for the midsummer precipitation anomaly in Shanxi is studied based on BCC_CSM output, NCEP/NCAR reanalysis data, reconstructed sea surface temperature (SST) and station precipitation in Shanxi during 1990-2017. The typical anomalous patterns of midsummer precipitation in Shanxi are analyzed, and then predictors associated with SST from BCC_CSM output are identified in term of statistical significance to the typical anomalous patterns of midsummer precipitation in Shanxi. The multifactor stepwise regression is used in statistical downscaling prediction. The improvement of prediction skills in downscaling results are apparent, as measured by the temporal and spatial anomaly correlation coefficient (TCC and ACC), the prediction consistency of the anomaly sign (PC), and the prediction score (PS) between hindcasts and observations. TCC from downscaling results crosses the 95% significance threshold in most parts of Shanxi and exhibits 99% confidence level in the middlesouthern Shanxi, and the simulated precipitation from BCC_CSM shows too small value to reach statistical significance in Shanxi. ACC increases from -0.02 for BCC_CSM to 0.35 for downscaling results, and the corresponding PC and PS are improved from 53.3% to 66.8% and from 65.6% to 78.9%, respectively. In the operational midsummer precipitation prediction in 2018 by using above downscaling method, ACC is 0.42 and PS is 70.8%.