Tolerance directly affects the performance and cost of the mechanical product. Tolerance analysis is a very useful approach for evaluating the accumulation of uncertainties caused by individual part tolerances. Worst Case (W-C) method and statistical methods are commonly used tolerance analysis methods. However, the result of W-C method is overly pessimistic, and the statistical methods adopt idealized distribution assumptions. In this paper, a novel C-NPS method combing Convex Method and non-probabilistic set theory (NPS) is put forward to address the above tolerance analysis problem. In this method, uncertainties of both part tolerances and assembly variations are modeled using NPS, then these part uncertainties are accumulated together to calculate the assembly function using Convex Method. Thus, the variation caused by each feature in the mechanical assembly can be estimated. C-NPS method is more suitable for tolerance analysis of different tolerances when the tolerance probability distributions are unavailable. The application of the method is illustrated through a one-way clutch mechanism assembly problem, and the advantages of this method are presented. The proposed method can be regarded as an attractive supplement to the tolerance analysis field.