The aim of this study was to evaluate the optimization of three-dimensional steel moment frames structures with torsional irregularity, via particle swam algorithm under earthquake load. In this way, to produce torsional irregularity, the center of stiffness has been kept constant with symmetrical grouping of the members, and the center of mass was placed with specific distances from the center of stiffness. Besides that, two types of three-dimensional steel moment frames structures have been modeled. The first structure has box-shaped columns which is assessed with 0 Percent, 20%, 40% and 60% torsional irregularity. The second structure includes cross-shaped columns which is evaluated with the same condition as first structure. Particle Swarm Optimization (PSO) algorithm has been used to achieve the best structure in terms of: weight, maximum drift, dimensional fit of joints and strong column to weak beam. In addition, in the optimizing process of elements, the required strength of the sections according to AISC 360-16 is satisfied under the LRFD method. The AISC360-16 database is also used for sections of structures (I-shaped beams, box-shaped and cross-shaped columns). The optimization results demonstrated that the weight of the first structure (box-shaped column) with, 20%, 40% and 60% torsional irregularity was higher in comparison with the regular optimal state with 43.242 ton, to 5.5, 9.9 and 11.3%, respectively. Moreover, the weight of the second structure (cross-shaped column) in same noted condition was 0.31, 9.4 and 11.9% more than the weight of the structure in regular optimal state with 54.915 ton, respectively.