The frequency and amplitude characteristics derived from the Ground Penetrating Radar (GPR) data have been widely applied to object recognition in karst areas, but still meet some limitations. Here we present a new method using wavelet energy spectral analysis. First we analyze the GPR signals of typical samples in karst areas and obtain their wavelet energy spectra, which consist of the energy eigenvectors on different scales and frequency bands. Then the object recognition is achieved by comparing the characteristic energy spectra with those of studied object. Both the data analysis and experiments demonstrate that the wavelet energy spectrum can directly show characteristics of object signals, which is very effective for the object recognition in karst regions from the GPR survey.