The number of fatigue test data is generally small, thus under the assumption of weibull distribution, the statistical confidence of characteristic life should be taken into consideration. In this paper, probability estimation methods of characteristic life are summarized, such as Bayesian method and Bootstrap method of virtually expanded sample. Then seven groups of fatigue test data for straight lugs are applied for comparison of these methods. Because EDF fitting methods can’t offer substantial guidance to the rationality of characteristic life’s estimations, a new method is proposed that all the test data should fall into (1- a) percentile interval of the life distribution. Through comparison and analysis, it can be concluded that estimations by Bayesian method are smaller than that of Bootstrap method, and the proposed judgment method is effective.