The deformation capacity of steel structures subjected to seismic or other repetitive loadings depends on structural configuration, member and connection characteristics, as well as on the cyclic response of material. While monotonic response of steel is generally available from standard tensile tests, information on response of steel under low cycle and extremely low cycle loading is limited. For numerical analysis, a basic option to determine the steel structures response is to use the tensile properties affected by a calibrated damage of the material. The literature offers several models for more rigorous analysis but these models are more reliable if their calibration is based on experimental results. Experimental results are found in the literature but mostly for low strain cycles. In the case of higher strains, the testing procedure, including the testing machine, requires supplementary support to avoid lateral displacement of the machine cross-head which makes these tests more difficult to be performed. A dedicated device for the cyclic loading was designed against lateral displacement of the machine cross-head. In comparison to the tensile test specimen, the cyclic test specimens have a shorter calibrated length. The paper presents the results of an experimental research on four European steel grades (S275, S355, S460 and S690) subjected to cyclic loading with constant and variable amplitudes. The specimens were tested with a universal testing machine of 250 kN under quasi-static strain rate. The results emphasize the stress-strain curve, number of maximum cycles for constant amplitude, maximum strain for variable amplitude and hardening or softening for the steel response to cyclic loading. Based on experimental data, parameters for modelling cyclic response of steel in Abaqus were determined.