Celestial spectra should be preprocessed before automated classification to eliminate the disturbance of noise,unstableness of observation environment,and flux aber- rance,etc.In this work,methods for data representation and flux normalization are studied. First,the necessity of proper data representation and flux normalization is clarified by an- alyzing their influence on spectra automated analysis.Second,a flux varying model is pro- posed and several different normalization schemes are assessed.While on galaxy and quasar classification,experiment results show that the log-wavelength representation performs bet- ter than the wavelength representation,and surprisingly the commonly used normalization scheme performs worse than others,for example,a median-based one.