A laser-induced breakdown spectroscopy (LIBS) technique based on artificial neural networks (ANN) is proposed for high accuracy elemental quantitative analysis. A combination method of laser induced breakdown spectroscopy with an artificial neural networks is employed to predict the concentrations of Cr and Ba in soil samples. A back-propagation algorithm with momentum coefficient and adaptive learning rate is used and served as a calibration strategy for LIBS. The quantitative results and relative standard deviation of repeated predictions are obtained. The results are compared with those obtained by conventional calibration curve methods. The results presented demonstrate that the combination method of LIBS with ANN performs better than conventional calibration curve methods in quantitative detection of Cr and Ba in soil with improved accuracy and measurement precision in terms of relative standard deviation. Furthermore, it is an excellent method for LIBS quantitative detection for heavy metal in soils.