In order to construct a high performance manufacturing line in a short time, it is inevitable to simulate manufacturing process from the view point of accuracy. This paper proposes a method of evaluating the accuracy of assembled product based on a geometric model. Machining errors, assembly errors and measuring errors are modeled as probability distribution functions in Monte Carlo method. Selective assembly is simulated, and the number of classification is evaluated. The result indicates the number of the classes improves the assembly accuracy, but decreases slightly the efficiency of the total assembly line.