To address the state-space explosion and probability resource mismatch caused by the large node scale of the composite aircraft Electrical Return Network (ERN), a reliability assessment method combining topological screening and probability guidance is proposed. Based on the percolation theory and the Molloy-Reed criterion, a truncation boundary is constructed to identify and eliminate physically disconnected states through network degree-moment characteristics. Furthermore, a probability-guided model is established by integrating dynamic risk weights and a posterior consequence truncation mechanism, which guides the Monte Carlo sampling to focus on high-risk failure regions. The results of the 82-node primary case and the 29-node comparison network show that the proposed method reduces computational overhead while ensuring assessment accuracy. Specifically, in the 82-node primary case, the number of evaluations is reduced by 96.25% and the computational efficiency is improved by 18.25 times compared with the traditional Monte Carlo method, demonstrating good applicability across different system scales. In addition, the sensitivity analysis of heterogeneous component life parameters indicates that the degradation characteristics of different component types have different effects on the degradation process of system connectivity reliability. The proposed method can provide a feasible assessment framework for the full-life-cycle connectivity reliability analysis of the composite aircraft ERN.