Abstract: The asymmetric cyclic load fatigue tests controlled by force of commercial pure titanium TA2 were carried out, and the fatigue life and fracture morphology of the pure titanium under different maximum stresses (245, 253,270,300 MPa) were studied, and the strain evolution and stress-strain response under asymmetric cyclic loads were analyzed. Based on the theory of continuous damage mechanics, the fatigue damage evolution prediction models based on ratchet strain and total strain energy density were established to predict the fatigue damage of the pure titanium, and the predicted results were compared with the test results. The results show that the fatigue life of the commercial pure titanium TA2 increased with the decrease of the maximum stress. When the maximum stress was 245, 253 MPa, the area of plastic ductile fault zone was small, fatigue crack growth zone appeared in the fracture, and the failure mechanism was fatigue crack and crack growth failure; the evolution of fatigue strain was composed of initial loading stage, stable stage and failure stage; with the increase of maximum stress, the plastic strain energy and the average strain corresponding to the same fatigue life increased. When the maximum stress was 270, 300 MPa, the fracture was mainly composed of fiber-like plastic ductile fracture zone, and the failure mechanism was cyclic plastic large deformation failure; the stability stage of fatigue strain evolution was not obvious; the plastic strain energy increased greatly with the increase of the maximum stress, and the average strain under the maximum stress of 270 MPa was larger when the cycle number reached 70% fatigue life. The fatigue damage evolution prediction model based on ratchet strain had a good prediction effect on damage values in the middle and late fatigue period (cycle number of no less than 40% fatigue life) with the maximum relative error no more than 12.5%. The relative error of damage value predicted by the fatigue damage evolution prediction model based on the total strain energy density was basically less than 23%, and the overall prediction stability was poor.