To evaluate the effect of split delivery and waiting time on Vehicle Routing Problem (VRP) with stochastic travel time, by considering the waiting time under soft time window, a stochastic programming model with correction was formulated to solve the Split Delivery VRP (SDVRP) with stochastic travel time. Meanwhile, an improved Particle Swarm Optimization (PSO) was proposed for this problem. At first, an improved relative position index method was used to decode the integer type code in which some customers (the split customers) may appear more than once. Then, an adaptive selection process was used to update the velocity in which the length of the involved vectors are different from each other besause of the split delivery. At last, the path relinking method was used to update the position of the swarm to deal with the information loss caused by the transformation between the continuous space and the discrete space. The experimental results on modified Solomon's instances show that the all-in cost is averagely increased by about 3%, and customers tend to choose split delivery when considering the waiting time. The use of split delivery can effectively reduce the all-in cost (2%) and the use of vehicles (0.6); in addition, in some instances, especially in the R2 instances, the waiting time can be reduced by about 0.78%.