During the process of dynamic analysis of the robot, the researchers generally ignored or not fully considered the effect of joint frictions on the dynamic model of the robot. Although the process of mathematical deduction was simplified, the model was quite different from the practice. Taking a 3-RPS parallel robot of which the rotation axes were parallel arranged as the research object, a dynamic modeling method based on modified Lagrange operator was presented. Firstly, the DOF of the moving platform was studied based on the screw theory and the kinematics model of the system was established by vector method. By analyzing the kinetic energy and potential energy of the whole moving components, the ideal dynamic model of the parallel robot was set up based on Lagrange equations. After that, taking the single chain, telescopic rod and moving platform as the research objects, the constraining forces of all motion joints were solved based on the D′ Alemberts principle, the joint frictions were regarded as the nonconservative forces of the system, based on the ‘Coulomb+viscous’ friction model, the work done by joint frictions was disposed accurately and quantitatively. Finally, the work done by joint frictions was always negative, some mechanical energy of the parallel robot was converted to other forms of energy and the mechanical energy of the system was reduced. Based on the ideal Lagrange operator，the Lagrange operator was amended based on the form of negative work done by frictions, and the system dynamic model considering all joint frictions was established. At the same time, the dynamic models with and without considering joint frictions were simulated and contrasted. In a simulation cycle, the results showed that the relative errors of the first, second and third driving forces were 18.1%, 12.6% and 16.5%, the results provided a theoretical basis for the frictional compensation of the robot control system and the analysis process also had reference significance for other parallel robots performance analysis and optimization.