This study presents a rapid, noninvasive method for screening of ocular diseases using tear Raman spectroscopy combined with different classification algorithms. Tear samples from 69 patients with ocular disease (including 14 cases of conjunctivitis tears, 17 cases of blepharitis tears, 10 cases of meibomian gland cyst tears, 18 cases of cataract tears, and 10 cases of glaucoma tears) and 48 healthy volunteers were measured in this experiment. In the measured tear spectrum, the tentative assignment of Raman peaks indicates specific biomolecular changes between the groups. First, partial least squares (PLS) and principal component analysis were used to extract features and greatly reduce the dimension of the spectral data. Then support vector machine (SVM) and back-propagation neural network algorithms were used to establish a discriminant diagnosis model. Finally, the best diagnostic results were obtained from the PLS-SVM model with accuracy, specificity, and sensitivity of 85.71%, 100%, and 76.19%, respectively. Our results suggest that tear Raman spectroscopy combined with multivariate statistical methods has great potential for screening for ocular diseases.