Portfolio optimization problems are formulated as a quadratic integer program. The formulated model can handle discrete assets, transaction costs, and logical constraints. Although many optimization methods have been proposed to solve portfolio optimization problems, they usually can not guarantee to identify a global optimum but easily be trapped into a local optimum solution or a feasible solution. This study proposes an efficient global optimization method to find a global optimum of a portfolio optimization problem. A nonlinear portfolio selection model is transformed to a linear model with fewer binary variables and constraints. Comparing to the current methods, the proposed method can effectively improve the computational efficiency. Furthermore, this study utilizes some numerical examples to illustrate the effectiveness of the proposed method.