Abstract: In recent years, the environmental pollution has become more and more serious, plagues of pests happen in many areas, which arouse great concern about the safety of agricultural products. In order to quantitatively predict the demand of pesticide, this paper, based on the prediction model, established a residual ARMA model to adjust the predictive model. First, a regression model was established, then identified whether the residuals belonged to the white noise, and established the ARMA model for non-white noise residuals. The results showed that, the mean square residual of single regression model was 25.39, the mean square residual of the combination of ARMA model and regression model was 13.79, reduced by 11.60 and the prediction accuracy was improved. The model was significant, the parameters were very appropriate. What’s more, this model’s mean square residual was reduced by 17.73 compared with the Logistic model, effectively improved the prediction accuracy. From 2012 to 2014, the amounts of pesticide forecasts are 181.068×10⁴ t, 183.64×10⁴ t and 187.96×10⁴ t.