<p>The probabilistic prediction of fatigue damage accumulation is of great importance for engineering structural components subjected to variable amplitude loading conditions. A stochastic continuum damage mechanics approach implementing the Sobol’ global sensitivity analysis is developed to predict the scattering of remaining fatigue life (RFL) under two-step amplitude loading (TSAL). This paper focuses specifically on the investigation of how the uncertainty of RFL is apportioned by uncertain model parameters, and based on the obtained Sobol’ results, the underlying fatigue processes behind load sequence effects are assessed. The proposed approach is validated with experimental data under TSAL considering loading sequence scenarios and cumulative cycle ratio values in the first-step loading. It is found that the model parameters that contribute most to the prediction of RFL are the number of cycles to failure at the first stress level and/or the number of cycles to failure at the second stress level of TSAL tests regardless of the load sequence scenario. It is further revealed that the computed Sobol’ results can reasonably be used to explain the underlying fatigue processes that are active at different stress levels.</p>