This work proposes a formulation for the optimum design of columns of reinforced concrete tall buildings. The design variables are the dimensions of the cross-sections and the amount of longitudinal reinforcement in the columns and the concrete characteristic resistance. Aiming at reducing the size of this class of problem, the multilevel optimization technique is applied so that the problem is subdivided into a global problem connected with a series of dependent subproblems. In this kind of problem, the design variables are subdivided into two groups, namely, the group of the so called global variables, represented here by the cross-section dimensions of the columns and by the concrete characteristic resistance, and the group of local variables, represented here by the areas of the longitudinal steel reinforcement in the cross-sections of the columns. Consequently, a global optimization problem is defined for the global variables, at the global level, and local independent optimization sub- problems are defined for the local design variables, at the local levels, following each iteration at the global level. The objective function of the optimization problem is the total cost of the columns of the building. The buildings are modeled as three dimensional frames and geometric nonlinear behavior is considered in the structural analysis. The assumption of rigid diaphragm is adopted for the structural behavior of the slabs and the torsional stiffness of elements is neglected. Dead, live and wind loads are applied to the model. Constraints related to ultimate and serviceability limit states, as well as lateral constraints, are imposed to the design variables. The limit states are defined according to the Brazilian Code for reinforced concrete structures NBR-6118 (2003). The solution of the optimization problem is obtained using Mathematical Programming techniques.