Abstract: The distribution density of the driver's start-reaction time was estimated using non-parametric kernel density to accurately analyze the driver's start-reaction time. Gaussian kernel was chosen as the kernel function and the optimal window width was obtained by the recursive method. The estimation results of kernel density were compared with the results of normal distribution and lognormal distribution by distribution fitting and hypothesis testing. The results show that the nonparametric kernel density estimation of the driver's start-reaction time is accurate and effective. The method overcomes the problems of the unknown prior distribution type. The curve of kernel density estimation is more intuitive to see the changes of driver's start-reaction time in each time period and the overall distribution pattern.