A Bayesian reliability assessment method of small sample failure data for numerically controlled machine tools with imperfect repair was proposed. The selection of prior and posterior distributions for virtual age model of generalized renewal process was discussed. Bayesian point and interval estimates of model parameters and reliability indices including cumulative failure number, failure intensity and reliability etc. were obtained via Markov chain Monte Carlo (MCMC) sampling. A real example was analyzed, whose results show that Bayesian reliability analysis method has a higher accuracy than maximum likelihood estimate because of its incorporating prior information.