This paper aims to improve the predictability of extreme droughts in China by identifying their relationship with atmospheric teleconnection patterns (ATPs). Firstly, a core drought region (CDR) is defined based on historical drought analysis to investigate possible prediction methods. Through the investigation of the spatial-temporal characteristics of spring drought using a modified Mann–Kendall test, the CDR is found to be under a decadal drying trend. Using principal component analysis, four principal components (PCs), which explain 97% of the total variance, are chosen out of eight teleconnection indices. The tree-based model reveals that PC1 and PC2 can be divided into three groups, in which extreme spring drought (ESD) frequency differs significantly. The results of Poisson regression on ESD and PCs showed good predictive performance with R-squared value larger than 0.8. Furthermore, the results of applying the neural networks for PCs showed a significant improvement in the issue of under-estimation of the upper quartile group in ESD, with a high coefficient of determination of 0.91. This study identified PCs of large-scale ATPs that are candidate parameters for ESD prediction in the CDR. We expect that our findings can be helpful in undertaking mitigation measures for ESD in China.