Experimental studies were carried out to investigate the ratcheting behaviour of three high strength rail steels of similar nominal hardness but with different chemical compositions subjected to uniaxial and non-proportionally bi-axial compression–torsion cyclic loading conditions. Different axial stress and equivalent shear stress amplitudes and different non-proportional loading paths were considered. Experimental results show that an obvious cyclic softening (i.e., the stress amplitude decreases with the increase of cyclic number) occurs in all three steels under uniaxial strain cycling. The ratcheting strain and ratcheting strain rate increase with the axial stress and the equivalent shear stress amplitudes under bi-axial compression–torsion stress cycling. Moreover, both ratcheting strain and ratcheting strain rate are strongly influenced by the non-proportional loading path. Among the three rail steels, it is found that the low alloy heat-treated rail steel grade has a better resistance to ratcheting than the two hypereutectoid rail steel grades. The hypereutectoid rail steel grade with a higher carbon content gives a lower ratcheting strain and a lower ratcheting strain rate than the hypereutectoid rail steel grade with a lower carbon content under higher loading amplitude. To simulate the ratcheting behaviour of the high strength rail steels, an existing cyclic plasticity model was modified by coupling a non-proportionally multi-axial parameter into isotropic softening and kinematic hardening rules. The method to calibrate the material parameters for the plasticity model and the simulated results validated with experimental data for the three studied rail steels are presented in the paper. This modified plasticity model with the calibrated material data from the experimental study can be applied to investigate the ratcheting behaviour of the three rail steels under wheel–rail cyclic rolling contact in practice.