Because of its importance in many industries, topology optimization of structures has recently been the focus of many researches. Various approaches have so far been proposed in the development of the method. The reason for such diversity lies in the fact that the method inherently suffers from some instability effects. One of the effective approaches for eliminating such unwanted instabilities is sequential refinement of the mesh used for the design variables. In this thesis, the capability of the approach is investigated in topology optimization of structures under dynamic transient loadings. The results are compared with those obtained by another approach using filtering technique. A node-based as well as an element-based definition of the design variables is considered for the studies. Two types of structures are considered for topology optimization; in the first one the primary multi-story structures are modeled as mass-less plates with concentrated mass at the story levels and in the second one the primary structures are considered as plates with distributed mass. The multi-story structures are subjected to a base excitation induced by an earthquake.The plates with distributed mass are modeled as cantilever beams and are subjected to concentrated transient loads at the tip. Steel, as well as another stiffer material, is considered for the plate material. In addition to the well-known earthquake records, records with sinusoidal variation as well as those with Gaussian wave packets are considered for excitation of the structures. The effect of damping on the final topologies has also been investigated. Due to the lack of rich literature for suitable objective functions, an appropriate objective function is proposed and tested through several numerical examples. In order to reduce the computational cost and time, a new method based on construction of an artificial record from the original one, is also proposed. To this end, a series of Gaussian wave packets are used to define a record having frequency content similar to that of the original one. It has been shown that the topologies obtained from both records, i.e. the artificial and original ones, are similar. It has been experienced that with such a record replacement the computational time decreases significantly. Finally, a new method is proposed for topology optimization of structures composed of more than two materials. In this method, the number of variables remains unchanged when the number of materials is increased. An immediate result of such an effect is reduction of the optimization time. In several problems defined for structures under static and dynamic loading conditions, effectiveness and sufficiency of the method is shown. The obtained topologies indicate the robustness of the proposed method in multi-material problems.