Most of the methods in flexible rotor balancing currently used rely on the repeated running up and down processes of the balanced machine. The ordinary balancing methods without test runs are regularly based on too many hypotheses and depended on prior knowledge of bearings. The latter restricts their wide application in practice. This paper establishes a balancing process frame with both unbalance and bearing parameters as inputs and the vibration responses on bearings as outputs. The authors determine the rotor unbalance as an inverse engineering problem using genetic algorithms (GA). This method can simultaneously evaluate the bearing parameters and their working situations in the service environments of large rotating machinery.