By use of the BP neural network algorithm, the principal component analysis method was used to obtain the main factors affecting the pesticide residues including the relative molecular weight of pesticides, temperature, precipitation, pH, CEC, organic matter, application concentration and harvest interval. These factors were then used as input variables to preliminarily build the pesticide residue prediction model. The relative error of prediction was 0.92%-18.93%, the average relative error was 7.42%, and the absolute error was 0.001-0.153 mg/kg, and the coefficient of determination of BP neural network prediction model was 0.962 05. It can be seen that in the face of complex natural environment and citrus germplasm characteristics, the pesticide residue prediction system on citrus based on BP neural network showed a high prediction accuracy for the residues of various pesticides on citrus, indicating that it was feasible to apply machine learning algorithm to pesticide residues detection on citrus.