In order to reduce the influence of some factors such as the electromagnetic wave attenuation and noise jamming on detection effect of GPR, a predictive deconvolution method based on biorthogonal wavelet was proposed (short for PDBW method). In the PDBW method, the biorthogonal wavelet basis with the minimum reconstruction error for detection signal of GPR was chosen, using this wavelet to decompose the GPR signals into different frequency band sub-signals, and then predictive deconvolution and other filter method were used to deal with each frequency band sub-signal in time domain, at last the results of PDBW method by reconstructing each sub-signal were get. Using the PDBW method to deal with the detection signal of experiment, the results show that, comparing with the predictive deconvolution, the PDBW method could restrain the multiple echo interference, identify the deep signal precisely, and enhance the signal to noise ratio of deep signal, thus improving the detecting resolution and the accuracy of image analysis of GPR signal.