A method for early diagnosis of keratoconus based on multi-modal data considers a mutual relationship between both eyes using four refractive maps for corneas of the both eyes and absolute corneal elevation data, and combines the deep convolutional network method, the traditional support vector machine (SVM) method in machine learning, and the elevation map enhancement method with adjustable best-fit-sphere (BFS) to identify sensitivity and specificity of a focus and balance the sensitivity and specificity. With multi-dimensional comprehensive judgment of a keratoconus morbidity with a patient as a unit, combined with binocular data including both manual selection features and deep network learning from big data, the diagnosis method has higher robustness and accuracy.