Nd:YAG laser, OceanHR2000+containing ICCD spectrometer are used for the laer induced-breakdown spectroscopy (LIBS) system. LIBS spectral data is acquired after the reasonable experimental parameters are setting. Using back-propagation artificial neural network (BP-ANN) algorithm, the heavy metal samples of Cr in soil which contain different concentrations of Cr are analysed. The peak at 425.435 nm of Cr is chosen to be the inputs whose intensity is normalized by the inner reference element of Fe at 425.079 nm. The net is trained and the qualitative analysis model is construted. Using this model sets of data are used for validation. The results of relative standard deviation (RSD) and mean square error (MSE) are both better than the traditional method of calibration curve.