Based on Bayesian inference, parameters and their validity of four queuing models with a single server in Markov and non-Markov system were estimated, and face-to-face currency exchange data of a bank was tested. By using approximate Bayesian computation, it is effective to solve the difficulty in analytically expressing the likelihood function in complex queuing models. Based on R package, parameters’ estimations and posterior distributions of arrival time and service time for four queuing models: M/M/1, M/G/1, G/M/1, GI/GI/1 were given. For M/M/1: estimated value (true value) is 1.02(1)、1.12(1/0.9) respectively, which are close to the true ones; for M/G/1: they are 1.17(1)、1.21(1.2) and estimated parameter of service time is superior to that of arrival time; for G/M/1: they are 0.49(0.5)、0.96(1) and estimated parameter of service time is more accurate from the posterior distribution; for GI/GI/1: they are 1.16(1.1)、1.07(1)、0.23(0.251)、0.22(0.25) and estimated value of each component is relatively close to the true value. Furthermore, for real data, distribution of fitting data of estimated parameters is close to that of Ausin’s fitting data. The results show great efficiency of parameter estimation based on approximate Bayesian computation in queuing theory and great application to the real world.