The thalassemias are a group of anemias result from inherited defects in the production of hemoglobin. The current techniques for screening and diagnosis of thalassemia are time consuming and complex. A laser tweezers Raman spectroscopy (LTRS) setup was used to trap single erythrocyte from patients with thalassemias and normal donors, and to collect the Raman scatting of trapped cell. Blood samples obtained from 11 patients with non-deletional HbH disease (HbH-CS), 11 patients with β-thalassemia major, and 11 normal controls, were tested. Principal component analysis (PCA) algorithm combined with back-propagation neural network predictive model was performed to distinguish abnormal erythrocyte. The PCA results reveale that the difference between normal controls and HbH-CSs is significant with the predictive accuracy of BP network as high as 97.90 %. The predictive accuracy between normal controls and β-thalassemias major is 90.72 %, and 86.28 % between HbH-CSs and β-thalassemias major. These results tally closely with the corresponding averaged Raman spectra. Under different experimental condition, the predictive accuracy showes similar results. This pilot study can serve as a useful probe for developing a rapid, simple, reagent-free method for distinguishing of thalassemia erythrocytes.