This paper proposes a substructure-based distributed-collaborative surrogate modeling approach for improving the accuracy and efficiency in the estimation of the creep-fatigue reliability of gas turbine blades with the integration of the Least Absolute Shrinkage and Selection Operator (LASSO), kriging, Distributed Collaborative (DC) strategy and substructure simulation method, called as the substructure-based DCLKM. Further, the creep-fatigue reliability assessment framework is set by combining the proposed substructure-based DCLKM with some theoretical models, which takes into account the uncertainties of structural sizes, applied loads and material properties and the nonlinearity of creep/fatigue damage interaction. Firstly, the total strain range, mean stress, absolute temperature and maximum stress at the critical location are regarded as the first-level output responses, and their required samples are predicted through the fitted surrogate models. Secondly, the creep life and Low-Cycle Fatigue (LCF) life are regarded as the second-level output responses and predicted. Finally, the creep-fatigue damage and reliability assessment are performed for several given applied cycles, respectively. The proposed substructure-based DCLKM is validated on a case study. The results show that the proposed substructure-based DCLKM is feasible with high accuracy and efficiency in the creep-fatigue reliability assessment of turbine blades.