A tandem AGV system is based on partitioning all the stations into non-overlapping single vehicle closed loops with additional stations provided as an interface between adjacent loops. For an efficient use of this configuration, it is required to solve the load routing problem(LRP), which is primarily based on the fact that a load may be handled by several vehicles and moved through several loops before it reaches its destination. In this paper, a heuristic based on genetic algorithm(GA) is first developed to solve LRP. The first model obtains the optimal route of each job and the optimal direction of each loop when the vehicle in each loop travels unidirectionally. The second GA model obtaines the optimal polling sequence of the empty vehicle in each loop, when the vehicle can move bidirectionally.