This paper addresses the problem of frequency offsets and channel gains estimation for MIMO system in flat-fading channels. Based on the MUSIC (Multiple Signal Classification) and the ML (maximum likelihood) methods, a new joint estimation algorithm of frequency offsets and channel gains is proposed. The new algorithm has three steps. A subset of frequency offsets is first estimated by the MUSIC algorithm. Then all frequency offsets in the subset are identified by the ML method. Finally channel gains are estimated by the ML estimator. The algorithm is a one-dimensional search scheme and therefore greatly decreases the complexity of the joint ML estimation, which is essentially a multi-dimensional search scheme. The performance of this algorithm is evaluated by Monte Carlo simulations.