The invention relates to a modeling method for a unit combination problem of a power system. The method comprises the steps of determining thresholds for classifying units according to capacities; classifying the units according to the thresholds; respectively determining the number of unit types for optimization and the unit number of each type; establishing a unit combination mathematic model; and determining the number of the starting units of each type, the total output of the units, and the start and stop schedule and the output schedule of the units. According to the method, the conventional units of the power system are classified according to the installed capacities, and the unit combination mathematic model is established by using the number of the conventional units of each type as an optimization variable. Compared with the traditional method of establishing the mathematic model by using the start and stop states of the individual unit as optimization variables, the method can greatly reduce the number of the variables of the mathematic model, and is effectively improved in the speed of solving the unit combination problem of the power system.