The next generation wireless systems, like International Mobile Telecommunications-Advanced (IMT-A), aim to provide high data rates, requiring spectrum allocation in the range of 100 MHz. Nowadays, we have both a limited available spectrum and inefficieny in its usage. It will be impossible to allocate 100 MHz to each operator in a same geographical area. Therefore there is a need for new spectrum access protocols to share the spectrum among different operators in a fair and efficient way. A game theory solution to spectrum sharing is discussed in this thesis. We proposed a dynamic game model based on regret lerning test to solve the spectrum sharing problem. We show through realistic simulations that it is possible to share the spectrum in a fair and efficient way.