The images of flue-cured tobacco leaves at different leaf positions were collected and preprocessed, to extract 5 absolute shape features of leaf length, leaf width, area, perimeter and minimum external rectangular area, and to calculate 5 relative shape features of narrowness, rectangularity, roundness, percentage of leaf width at the maximum and the angle between leaf width axis and center of mass. The feature vector was filtered out by the main component analysis to construct 5 tobacco leaf position recognition models based on K-nearest neighbor(KNN), logistic regression(LR), support vector machine(SVM) with linear kernel function and radial basis kernel function, and BP neural network, respectively. The recognition effects were compared for the five models. The results showed that the morphological feature parameters extracted based on the image contour features could reflect the characteristics of roasted tobacco positions more effectively. It could be seen that the BP neural network-based model has the best recognition effect with a recognition accuracy of 93.75% among the five recognition models, and the model decision coefficient is above 90% for both the training and test sets.