Abstract: Given the difficulty traditional phenomenological strength theories face in accurately determining failure boundaries, this study investigated C/SiC composites and their typical structures through a combination of numerical simulations and mechanical testing. By incorporating uncertainty quantification methods, a comprehensive analysis approach was proposed that establishes a probabilistic failure criterion for C/SiC composites—accounting for inherent material scatter—and evaluates structural reliability. First, mechanical tests were conducted, and the least-squares method was used to calibrate the tensor coefficients of the Tsai-Wu criterion; through the multi-source fusion of test data obtained under various conditions, material strength uncertainty was transformed into a statistical distribution of failure coefficients, thereby establishing the probabilistic failure criterion. Next, using a C/SiC composite box-type structure as the subject, the probabilistic failure criterion was applied to structural reliability analysis via Monte Carlo simulation. Finally, local sensitivity analysis was employed to quantify the contribution of various mechanical parameters to the structural failure probability. This method enables the prediction of material failure probability under arbitrary stress states and identifies the sensitivity of critical mechanical parameters to overall structural failure; it provides a valuable reference for predicting the damage of C/SiC composites under arbitrary loading and offers insights for the precise design of safety margins and the monitoring of structural load-bearing capacity.