This paper proposes a method to estimate the model order based on the singular values of a synthetic real matrix. The synthetic data matrix includes the complex observation data and its conjugate data, which fully exploits the existing data information, is more suitable for model order estimation of multiband signal fusion spectrum than the complex observation matrix. Simulation experiments show that this method has better results than the traditional model order estimation methods based on complex observation data matrix, the model order can be estimated at a lower signal to noise ratio (SNR).