Abstarct: In most of the structural optimization problems, all variables have been considered as deterministic and their uncertainties have been neglected. Recently, the reliability theory has been employed in the structural optimization by designers. However, reliability analysis of the structural systems is a very time consuming process and hence the structural system failure probability in many studies has been estimated in a conservative manner directly from the failure probability of the members to reduce the computation time, resulting in inefficient and uneconomical solutions. The aim of this thesis is to optimize the 3D steel trusses, independent of any design codes, taking into account the structural system, correlation between random variables and their real statistical distribution, such that the truss weight becomes minimum under a given level of safety. This type of design is known as the 4th level of the reliability theory in which the common optimization constraints such as allowable stress of members are replaced by the structural reliability. In this study, in order to increase the speed of the optimization and the reliability analysis, an improved distributed genetic algorithm, a parallel processing system, a developed algebraic force method and a multi-agent system baxsed on the artificial intelligence techniques have been utilized. Also, by using the probability concepts and the reliability theory, the Branch and Bound method baxsed on the failure paths has been developed. The optimization of truss structures under constraint of structural system reliability, without simplifying assumptions is a time consuming process and therefore impractical. In this study, using the proposed strategies, such an optimization has been carried out in a reasonable calculation time.