The detection model of sucrose content in tea was established by using near-infrared spectroscopy technology and a back propagation neural network algorithm. The quality of model prediction is improved by introducing a genetic algorithm. The prediction model was established using Fourier transform diffuse reflectance spectroscopy data of 120 tea samples mixed with sucrose. The prediction results of another 42 samples show that the correlation coefficient based on the traditional back-propagation neural network algorithm model is 0.7380. The root mean square error of prediction is 3.0754. The prediction accuracy is 83.3%. The correlation coefficient increased to 0.9419 after the introduction of the genetic algorithm. The root mean square error of prediction is 1.3176. The prediction accuracy is 88,1%. The training error is reduced by more than an order of magnitude. The experimental results show that the back-propagation neural network model can be used to detect the sucrose content in tea. Simultaneously, the genetic algorithm is introduced to optimize the initial weights and thresholds of the neural network to make the prediction error smaller.