Leaf mold is a common disease on tomato leaves, which seriously affects the quality and yield of tomatoes. In order to use hyperspectral technology to achieve early detection of leaf mold, 200 samples were first collected and the hyperspectral data of all leaf samples in the band of 927 to 1684 nm were obtained. According to the lesion area, all leaf samples were divided into 4 grades, and then the four pretreatment effects were compared, and the Savitzky-Golay convolution smoothing method was selected as the pretreatment method. Competitive adaptive reweighted sampling (CARS), iteratively retains informative variables (IRIV) and a combination of the two algorithms are used to select feature variables to establish tomato leaf mold hyperspectral support vector machine (SVM), CARS-SVM, IRIV-SVM, CARS-IRIV-SVM detection model. The results show that the detection accuracy of the four models for level 1 samples are all higher than 80.9%, and the recognition effect is good. The overall prediction accuracy rates of the four models are 79.41%, 86.76%, 85.29% and 92.65%. The CARS-IRIV-SVM model has the best results in identifying the characteristics of tomato leaf mold. The evaluation index of the model is Rp2=0.9103, RMSEP=0.138 and RPD=2.51, the prediction accuracy and overall accuracy of each level of the prediction set are 100%, 95.24%, 88.88%, 90.91% and 92.65%. The built model has high detection accuracy, indicating that the CARS-IRIV-SVM model based on hyperspectral technology is feasible for the classification and recognition of tomato leaf mold.