[Objective] Current micromechanical models pay limited attention to parameter uncertainty and interactions, which makes it difficult for their response results to reflect the dispersed nature of the properties of cement-based materials. Therefore, it is necessary to explore an analysis method that can simultaneously capture the effects of multiple parameters and their interactions on the responses of micromechanical models of cement-based materials. [Methods] To address the discrete distribution of response results in existing micromechanical models and to identify and control the influencing factors causing this phenomenon, a multi-scale micromechanical model of cement-based materials was constructed in this study. Cement-based materials were divided into four scales: calcium silicate hydrate gel, cement paste, cement mortar, and concrete. Considering the mineral composition of cement phases, aggregates, and the ITZ, a multi-scale micromechanical model capable of accounting for the randomness of input parameters was proposed. Meanwhile, probabilistic methods were applied to the constructed micromechanical model, and global sensitivity analysis was employed to quantify the effects of input parameter uncertainty on the elastic modulus of cement-based materials. [Results] The results showed that the proposed model exhibited good applicability in simulating the relationship between elastic modulus and hydration degree of cement-based materials across multiple scales and showed good agreement with experimental results. The discreteness of the model response results mainly originated from the cross-scale propagation of input parameter uncertainty, indicating that uncertainty at the concrete scale incorporated the uncertainties of input parameters at the mortar and cement paste scales. The total-order sensitivity indices, ranked from largest to smallest, were the elastic modulus of sand and coarse aggregates, the volume fraction of sand and coarse aggregates, the elastic modulus of hydration products, the volume fraction of cement clinker, and the elastic modulus of cement clinker. To identify the dominant sources of uncertainty within the model framework, particular attention should be paid to the elastic modulus of sand and coarse aggregates, whereas the volume fraction and elastic modulus of cement clinker can be regarded as insensitive factors. [Conclusion] Screening the number of input parameters has important practical significance for reducing computational complexity and improving the efficiency of model response analysis.