There is large amount of noise in the ground penetrating radar (GPR) record, which causes the target signal to be distinguished with difficulty at times. In order to increase SNR of GPR sections, the validity of KL transform to remove noise in GPR record was discussed. On the basis of KL transform theory, the calculating program was written out in Matlab environment. The GPR modeling record of finite difference time domain method and real data were processed by KL transform, the results were analyzed and compared with original records. According to the analyzing results, KL transform suppress boundary reflection signals and corner diffraction signals of modeling data, and remove high frequency noise in GPR real data. After the transformation the effective signals are intensified, and resolution of GPR sections and inversion accuracy are improved.