The whole roller flatness meter is the development trend of the contact type flatness meter for cold rolling. Under the action of dead-weight and external load, the bending deformation of the detection roller will change the pre-tightening force of the sensors that are installed in the meter. The additional signals are generated, which are similar with sinusoidal waveforms. They will affect the accuracy of flatness detection by superposition with the effective signals. This paper studies the generation mechanism. The recognition and elimination method of this interference signal is also studied. Firstly, according to the experimental calibration and industrial testing results, the wave characteristics of the flatness signal and the interference signal are found and analyzed. The effective flatness waveform distributed along the wrapping angle is located on the peak and trough of the interference waveform distributed along the circumference. Secondly, the elastic theory is applied to calculate and analyze the stress and deformation caused by the beding of the flatness roller. It reveals that the periodic change of the pre-tightening force of the sensor is the cause of the interference sinusoidal waveform. Finally, based on the detection signal data outside the wrapping angle range, a mathematical model that has accurate and quick feature to identify and eliminate the interference signal waveform is formulated, which utilizes the curve fitting optimization method of minimum error. The industrial application show that the proposed method can effectively eliminate the beding signal and improve the effect of flatness detection and control by 1~ 2 I.