Due to the limited battery power and pesticide capacity, the plant protection UAV need return to the supply point frequently in the process of plant protection. With the work area increasing, more time would be spent on battery replacement, pesticide filling and round trips between each return point and the supply point. So an appropriate path with the optimal return points must be planned before starting the work, in order to minimize the total time and improve the efficiency of the plant protection. For the purpose, a research was conducted on the path planning method for the plant protection UAV. Firstly, aiming at building an environment model which could describe the working area, the grid method was selected to divide the working area into small grids with the initialized weights, which were depended on the working area’s size and shape. Secondly, the UAV was made to fly from the current grid to the adjacent one with the highest probability, which was calculated according to both the grids’ initialized weights and the heading direction of the UAV. Incentive coefficients were added to the weights of the grids located in the front, left rear and right rear of the UAV so that the parallel routes were followed which moved from one extreme of the working area to the other alternately and turned at the boundary. Then the quantity and position of the return point could be outputted by controlling the distance in the spraying mode. Thirdly, a mathematical model was established. The quantities of the return times in the artificial planned path and the unplanned path were taken as the upper and lower limits of the search space respectively. The distance of each flight in the spraying mode was chosen as the variable, and the dimensions of which were depended on the search space. The objective was to obtain the optimal return points with the minimum time in the non-praying mode. After that the gravitational search algorithm (GSA) was applied to solve the model. Based on the methods and processes above, a new path planning method was proposed. Then the method would output the planned path with return points automatically by inputting the data about the environment and the UAV such as the size of the working area, the direction of the crop row and the speed of the UAV. At last, for the test of the performance of the proposed path planning method, a 700m×100m working area with the irregular boundary was taken as an example for the path calculation. The path calculated by the proposed method was also compared with the artificial planned path and the unplanned path respectively, which showed the non-praying distance of the proposed method was reduced by 14% and 68%, while the non-praying time was reduced by 21% and 36%. Furthermore, a field experiment with the real UAV was used to test the proposed deviation rectification algorithm. Finally, the study indicated that the proposed method which could produce paths with less working time was a reasonable, feasible and useful solution for the path planning problem of the plant protection UAV.