According to the specific needs of the agricultural information service assessment, ant colony algorithm was adopted to optimize the traditional neural network to avoid its disadvantages of low convergence speed and being prone to fall into the minimum. Evaluation Index system of agricultural information service was built and the neural network model was designed. Learning and training were carried out using the sample data of agricultural information service system. The final evaluation result was obtained through learning of 56 sample data and training of 24 sample data. The result showed that iterations of the BP model optimized was significantly reduced, learning rate and stability were also improved. The model was able to evaluate scientifically and objectively the service of agriculture intelligence agencies.