The differences between the selected similar products and evaluation object will lead to the deviation of the expected results when the reliability prediction is carried out by the traditional similar comparison method. In order to solve this problem, a new method of reliability prediction based on “function-movement-action” structural decomposition was proposed. A “function-movement-meta-action units” hierarchical model was established for reliability prediction with steps by the means of establishing function layer by analysing the product's all functions, establishing movement layer by analysing components' movements that completed corresponding functions of function layer, and establishing meta-units layer by analysing meta-actions that completed corresponding components' movements of movement layer. The interval analytic hierarchy process (AHP) was introduced to determine the weights that reflected the influence degree of offspring motion units to parent motion units. And a mathematical model was constructed to predict parental motion units' reliability from offspring motion units. The factors were determined that affected similarity evaluation process, so as to construct interval comprehensive evaluation cloud model to evaluate similarity between meta-action units and mechanical structures, and the most similar mechanical structures were chosen as referential structures. And the similarity value was used that concluded from interval comprehensive evaluation cloud model to correct the reliability data of referential structures. Then the corrected reliability data were used to predict every meta-action units' reliability of the meta-units layer. The reliability prediction value of the unit of motion was substituted into the mathematical model, so the parental motion units' reliability were predicted from offspring motion units layer by layer, so as to obtain the whole product's reliability prediction value. At last, numerical control rotary table was taken as a case to illustrate the feasibility and accuracy by using the proposed method. The reliability prediction value was obtained and compared with the traditional method. The result shows that this new reliability prediction method has better feasibility and higher accuracy.