In the new information era, high rate information transmission and high reliability are the main features of a wireless communication system as well as the main success factors of its commercial expansion. Multiple Input Multiple Output Orthogonal Frequency Division Multiplexing (MIMO-OFDM) wideband wireless communication technology is considered a lot because of its high data rate and robustness against multipath fading characteristics of the channel. The main question in a MIMO-OFDM system is how to obtain channel status accurately. In the blind method, channel estimation is performed by statistical analysis of the channel and knowing some features of the transmitted signal. Channel estimation may also be based on a training sequence known to the receiver. In this approach, first a known training sequence of OFDM symbols is transmitted actual and the channel information is extracted before receiving the data packet. The direct generalizing of channel estimation methods in OFDM communication systems to MIMO-OFDM needs calculating the inverse of high dimensional matrices. Increasing the number of receiver and transmitter antennas, the complexity of computations is highly increased and its implementation gradually becomes impossible. Since MIMO-OFDM communication systems decompose the channel matrix into singular values to overcome the inter-carrier interference, to assign transmission power to sub channels optimally and also to do time-space coding optimally, it can be said that channel estimation by singular value decomposition is an important technique to employ all the abilities of communication systems based on MIMO-OFDM. In this thesis, channel estimation method of MIMO-OFDM communication systems is analyzed and a new method for selecting training sequences using Alamuti scheme is introduced to increase the performance of this method.