The effective information of Raman spectrum is easy to be weakened or even submerged by the background noises in complex environment. The configuration similarity comparing method (CSCM), which is based on the shape similarity of the information and the background spectrum, is proposed to eliminate the background noises. Exampled as experimental data, the CSCM principle is analyzed and the processing steps are introduced. By comparing with other two traditional methods, the CSCM recovers more Raman information and the baseline is flat, which is advantageous in further data analysis.