Structural steel stands as a highly-utilized engineering structural material. The safety performance of structures is subject to severe influence from the damage evolution of initial defects. To achieve a more precise prediction of the ultra-low-cycle fatigue life of structural steel, a novel ultra-low-cycle fatigue fracture model that takes defect density into account was formulated in this paper, drawing upon the Rice-Tracey model and the GTN model. Uniaxial tensile tests and cyclic loading tests were conducted on smooth round-bar specimens of Q235 steel. In conjunction with ABAQUS finite element analysis, the material performance parameters and mixed hardening parameters were successfully derived. Through uniaxial tensile tests and cyclic loading tests on notched round-bar specimens, coupled with ABAQUS finite element analysis, the toughness parameters of the void growth model (VGM) and the cyclic void growth model (CVGM) for Q235 steel were calibrated. To validate the correctness and accuracy of the new model, a comparison was made between the life prediction results of the CVGM and the newly-proposed model in this paper against the outcomes of cyclic loading tests on 9 groups of notched specimens. The results demonstrate that the prediction results of the CVGM and the ultra-low-cycle fracture model considering void density exhibit a relatively close resemblance to the experimental results. Furthermore, the newly-proposed model in this paper showcases higher accuracy, lower dispersion, and a reduced number of parameters requiring calibration. This indicates that the proposed ultra-low-cycle fatigue fracture model accounting for defect density is applicable for predicting the ultra-low-cycle fatigue fracture life of steel structures.