Rotor assembly is a core tache in the whole process of aero-engine manufacturing. Preventing out-of-tolerance of concentricity is one of the primary tasks. Conventional assembly approaches are based on a manual test with the dial indicator, depending on experience appraises, which lack systematic and quantitative precision design theory. As a result, two issues need to be solved: the modeling problem of complicated geometric variations in three-dimensions, as well as the abnormal distribution of ubiquitous actual deviations. This work attempts to propose a novel probabilistic approach for three-dimensional variation analysis in rotor assembly. Based on rotor’s revolving characteristics and multistage stacking process, Jacobian–Torsor model is adopted to establish the variation propagation, and Pearson distribution family is used to derive the probability density function, which can quickly determine the variation distribution pattern and efficiently perform statistical variation analysis. A real case of mechanical assemblies consisting of revolving axisymmetric components is concerned. The results show that the suggested method has a similar accuracy, but much higher efficiency than conventional methods. Calculations agree with the experimentations, and the probability distribution type of the part’s variation has an appreciable impact on the final assembly precision.