Membership correction is an important direction in the improvement of fuzzy c-means clustering algorithm. This type of improved algorithms introduce fuzzy threshold to correct membership value, which greatly speed up the algorithm convergence. However, the adaptive value of fuzzy threshold is always a difficult problem. To solve the problem, a method is presented to select the parameter of fuzzy threshold based on similarity relation and physical attraction between data and clustering centers. The monotonicity, convergence and robustness of the parameter selection formula are discussed to verify the effectiveness of this method. Simulation shows that the parameter selection method is effective, adaptive and robust, which has high application value to parameter selection of membership modified FCM algorithms.