A new up-and-down method is developed and a new estimate formula for the variance of sensitivity variables is proposed. This formula fully considers the effect of test conditions on the estimating accuracy, resulting in higher order of precision. In order to give an objective demonstration, the Monte Carlo comparative results between the present method and the Bruceton method is given, which shows that the estimating values by the present method are much more accurate than those by Bruceton method in all conditions. Even with small samples, the present method still could obtain a good estimate value. Thus the proposed method not only solves the problem with traditional methods but also reduced the number of experiment, resulting in great saving. Finally, an application for safety fatigue limits is given.