The experimental modal analysis is widely used to analyze the dynamic behaviours of a machine, which is based on the measurement of frequency response functions when the structure under stationary is excited by a hammer or an exciter. In many cases, the dynamic behaviours under stationary is more or less different from those under operation, which are more important consideration in dynamic design. Under operation only the response data are available so that the frequency response functions can not be measured. In this situation the response ratio method has been commonly used by simply calculating auto-cross spectra among the response data which yields inaccurate results due to contaminated noise or closely coupled modes. In this paper the principal component analysis that is the eigenvalue decomposition of auto-cross spectra matrix is utilized to overcome these problems. In addition the AR method is introduced in order to calculate the spectra even in transient state under operation instead of FFT. The dynamic behaviours when a motor bike gets over a bump is dealt with to show how the new method proposed can provide detail and accurate analysis on dynamic behaviours.