Abstract: Accurate seasonal forecasting of summer precipitation is critical for climate adaptation and disaster risk reduction in China, yet remains challenging due to the variability of the East Asian summer monsoon. Dynamical climate models serve as the foundation for operational seasonal prediction but are constrained by systematic biases and uncertainties in initial conditions. To address this challenge, the prediction skill of a multi-model ensemble (MME) comprising advanced seasonal forecast systems is evaluated for summer (June-August) precipitation and key ocean-atmosphere precursor signals over China during 1993-2024. The analysis utilizes hindcasting data from eight seasonal prediction models. To ensure temporal consistency, an equally weighted multi-model ensemble mean (MME-mean) is constructed from seven models (excluding the shorter-record BCC-CPSv3, which is assessed separately). The verification is based on gridded observational datasets (CN05.1, GPCP), ERA5 reanalysis data, and HadISST sea surface temperature. Evaluation is conducted using standard metrics, including temporal correlation coefficient, anomaly correlation coefficient, and prediction score.Results reveal substantial spatiotemporal heterogeneity in the MME-mean skill for summer precipitation prediction. Predictive skill exhibits significant interannual variability, with higher skill found over the mid-lower reaches of the Yangtze, North China, and western Northwest China. In contrast, prediction skill is markedly lower over Northeast and Southwest China. Forecasts initialized closest to the target season (0-month lead) generally show the highest skill, underscoring the critical role of initial conditions. The MME-mean demonstrates proficient and stable skill in predicting large-scale circulation, particularly within the tropical and subtropical regions. The temporal evolution of the Philippine Sea anticyclone is captured with significant skill across all lead times. Conversely, predictive skill for mid-latitude circulation anomalies over Eurasia degrades rapidly with increasing lead time. Prediction skill for global sea surface temperature (SST) fields is notably high. The MME-mean also partially alleviates the characteristic “spring predictability barrier” for ENSO evident in individual models. However, a focused analysis for the mid-lower reaches of the Yangtze indicates that even the best-performing individual model fails to accurately predict precipitation anomalies in several extreme years, particularly flood years, which is attributed to more complex antecedent SST patterns. In summary, the MME-mean approach significantly enhances the predictive skill for large-scale precursor signals associated with summer precipitation over China. However, substantial challenges remain in accurately predicting regional precipitation anomalies, particularly for extreme precipitation events. Future efforts should also prioritize exploiting probabilistic forecasting reliability and integrating physically-constrained machine learning techniques for bias correction.