As a typical arid/semi-arid area, summer rainfall in northwestern China (NWC) is critical for the ecological environment there. The NWC summer rainfall prediction using statistical and dynamical models shows very poor performance. The dynamical model NCEP CFSv2 shows very low prediction skill in NWC summer precipitation. NWC precipitation index (NWC PI) is defined as the regional mean summer precipitation over NWC. This study first investigates the soil moisture and SST (sea surface temperature) signals in the preceding months, and CFSv2 April predicted summer large-scale circulation anomalies significantly related with NWC PI during 1982–2016 (2011 missing). The April soil moisture in the upstream central Asia, and SST in the preceding winter over tropical Pacific, Atlantic and Indian Ocean are identified as predictor candidates in the observation. Besides, six April predicted large-scale circulation factors in CFSv2 are also hired. The potential model predictors are selected generally based on three criteria: where CFSv2 has significant prediction skill, and NWC PI is significantly correlated with the variable in both the observation and its counterpart in CFSv2 prediction. Hybrid models are then constructed based on different predictor combinations. The leave-one-out cross validation illustrates that the hybrid model using three predictors (April soil moisture in the upstream central Asia, April predicted zonal wind over the Maritime Continent, April predicted SLP over the Mediterranean Sea) presents best performance, with anomalies correlation coefficient–ACC of 0.65, percentage of the same sign–PERC 76.5%, relative error–RE 6.97%, and root mean square error–RMSE 0.344 mm/day for NWC PI. The April soil moisture in the upstream central Asia is linked with the summer water vapor flux over NWC, and April predicted zonal wind over the Maritime Continent is related with the large-scale circulation over NWC, while caution must be taken when April predicted SLP over the Mediterranean Sea is used as predictor of NWC PI.