An extraction method for na unbalanced signal based on EMD and instantaneous frequency estimation is presented in order to minimize the disadvantages of speed fluctuation in dynamic balancing measurement. In contrast to a traditional stationary signal analysis method, the proposed method obtained instantaneous frequency from the key phase signal by cubic spline interpolation, and built an unbalanced signal whose amplitude and phase were identified by a least square algorithm. In order to improve the accuracy of parameter identification, na EMD algorithm was used to filter vibration data before extracting sample data for least square estimation. Simulation and experimental results showed that this method can overcome the adverse effects of speed fluctuation and increase the accuracy and stability of unbalanced signal measurement, which is suitable for applications in engineering.