This paper proposes an new Genetic Algorithm(GA) approach for short-term thermal unit commitment. Unit commitment is a complex combinatorial optimization problem which is difficult to solve for large power systems. Up to now, the Lagrangian relaxation(LR) is considered the best way in dealing with large-scale unit commitment although it cannot guarantee the optimal solution. Recently, GA has been successfully applied to combinatorial optimization problem. However, GA is time-consuming since it requires binary encoding and decoding to represent each unit operation state and to compute the fitness function throughout GA procedures. This causes huge computation burden, making it difficult to apply to large-scale system. To realize high speed computation, a new genetic operations such as a few individuals, quick estimation and intelligent mutation operators are introduced. The proposed algorithm has been applied to the large-scale unit commitment problem, and the simulation results show that better solutions are obtained in reasonable computation time.