Track association is a prerequisite to track fusion in distributed multi-sensor/multi-target tracking.With the traditional method, the correct association rate seriously declines in a dense target environment.To solve this problem, this paper presents a new track association method. Traditional fuzzy membership is smoothed, and similarity among tracks is considered from a whole view. Thus a modified fuzzy membership is proposed. In addition, a training method is designed to obtain the parameters based on a priori sequential association couples. Simulation shows that the method can easily set parameters and has better performance as compared to the classical fuzzy double-threshold method, therefore more suitable to practical applications.