Agricultural drought is generally defined as a deficit in soil moisture, and can affect plant growth and crop yields. Accurate prediction of agricultural drought with sufficient lead time can aid agricultural planning and reduce losses in agricultural production. In this study, the meta-Gaussian model was employed to predict agricultural drought (standardized soil moisture index, or SSI) during spring and summer in China based on the initial soil moisture conditions, antecedent meteorological drought (standardized precipitation index, or SPI) and large-scale drivers (El Niño-Southern Oscillation, or ENSO). Monthly precipitation and soil moisture data from Global Land Data Assimilation System, version 2 (GLDAS-2.0) were used to compute the meteorological and agricultural drought indicators. The conditional distribution of agricultural drought given multiple predictors was used for the statistical prediction. The autocorrelation of agricultural drought, the correlation between meteorological drought and agricultural drought, and the correlation between ENSO and agricultural drought were first evaluated to understand the predictors from soil moisture persistence, drought propagation, and large-scale drivers. We then employed the conditional distribution to predict agricultural drought over the period from 1948 to 2014, in which contributions of antecedent meteorological drought and large-scale drivers to the prediction performance were evaluated. Results showed that the prediction method performed well in semi-arid and sub-humid regions during spring, but did not perform well in humid regions during summer. In addition, the incorporation of ENSO provided useful predictability for long lead time prediction in certain regions (with significant influence of ENSO). The results obtained from this study can provide useful information for early agricultural drought warning across China.