Summary: Correlation difference is an important index commonly used in medicine. The incidence of chronic diseases is usually fitted by Poisson distribution. The confidence interval of the correlation difference is constructed by using the saddle point approximation method. At the same time, it is compared with the traditional 4 confidence interval construction methods by using the Monte Carlo simulation. Finally, it is used for the actual data analysis. The results show that the saddle point approximation method is near the nominal level in most cases, and the interval length of saddle point approach is shorter when the difference of coverage is small; especially in small samples, the saddle point approach is the best. So saddle point approximation is a good method for constructing confidence intervals of statistics and can be generalized in various fields.