The surface vibration signals are widely used since they have much combustion information. However, for an Internal Combustion Engine (ICE), the measured surface vibration signals are difficult to utilize because they contain non in-cylinder pressure excitation response. The vibration response signals excited by the in-cylinder pressure excitation (ICPE) and the reciprocating inertia force excitation (RIFE) are overlapped in both time and frequency domain. That means they cannot be separated effectively by conventional signal processing method. In this paper, a new strategy to extract ICPE response from measured vibration signals by pattern recognition method is proposed. A model is established to describe the RIFE response. Then, the RIFE response could be predicted and subtracted directly from the measured vibration velocity signals. The processing results indicate that a fourth-order model and the data of initial compression stroke can reach satisfactory results. The impact of the speed fluctuation can be ignored. Meanwhile, the trends of the processed vibration velocity and the rate of in-cylinder pressure rise are similar around the crank angle of the peak pressure, which further supports the method.