For the standard particle swarm optimization algorithm is easy to appear slow convergence speed, emerge premature convergence and fall into local minimum point in the later evolution, a kind of localization algorithm based on cross particle swarm optimization for wireless sensor networks was presented to solve these problems. The approach mainly included three stages: sink node selection, measure distances amendment and unknown sensor node localization. By referring to the crossover operation of genetic algorithm idea, cross particle swarm optimization algorithm could increase the diversity of particles and reduce the distance measure error and the influence of anchor node number on localization result. The simulation experiment result showed that the stability and localization accuracy of the method proposed are better than those of the standard particle swarm optimization algorithm. Under the condition of same measure error and the equal number of anchor nodes, the new method was compared with the shuffled frog leaping algorithm. And the compared results are as follows: the maximum of localization errors are 1.3378m and 1.7473m, respectively; the minimum of localization errors are 0.2583 m and 0.5615m, respectively; the average localization errors are 0.6512m and 10447m, respectively. Results indicate that the method proposed is suitable for agriculture wireless sensor network localization.