A method based on the use of a spectral analyzer has been developed in order to identify and mitigate radio frequency interference (RFI) in microwave radiometry. This method has been tested with L-band CAROLS airborne data highly corrupted by interferences. RFI is a major limiting factor in passive microwave remote sensing. Although the 1.4-1.427 GHz bandwidth is protected, RFI sources close to these frequencies may still corrupt radiometer measurements. In order to reduce RFI bad effects on the brightness temperature measurements, a new instrument called spectral analyzer has been added to the CAROLS radiometer system. A post processing algorithm based on a selective filtering with the division of bandwidth in subbands is proposed. Two discriminant analysis based on the computation of kurtosis and Mahalanobis distance have been compared, evaluated and validated in order to separate accurately the RF interference with natural signal.