The present work is devoted to predicting the response of cantilever plates by expanding the vibration data at limited positions to the unmeasured regions with poor accessibility. The predictive algorithm is based on the mode superposition method (MSM) and the combination method of beam functions (CMOBF). Firstly, the vibration mode function (VMF) of the cantilever plate is deduced offline according to the CMOBF. According to the VMFs calculated offline and the responses measured at limited points, the mode coefficients are updated by the MSM in real time. Finally, the displacements or accelerations of target positions are predicted by the product of the mode matrix and the vector of mode coefficients. The predictive process is achieved in the absence of specific expression and distribution of external loads. Besides, different from traditional modal decomposition and expansion methods, the present work is not dependent on the finite element (FE) model or repeated iterations. The predictive process is convenient to carry out with a small computation time online. As some application examples, the predictive method is applied to the cantilever plates under concentrated and moving loads by simulations and experiments. Both the acceleration and displacement signals are taken into account and examined.