Since the number of fatigue crack propagation data is related to cost and time and the crack propagation data are generally finite, adopting median rank or average rank as the probability value of a test datum will result in significant error. On the basis of analyzing disadvantages of traditional methods to identify distribution functions, which begin with overall fitting effect, this paper proposed a method by applying numerical simulation and fuzzy linear regression. It can be used to determine the distribution function of crack propagation data. Through fuzzy linear regression, the result that crack propagation size under certain load cycles are in logarithmic normal distribution can be acquired. In the same way, the number of load cycles is also in logarithmic normal distribution when the crack propagates to a certain size.