A large number of farm lands are contaminated by heavy metals in the process of industrialization and urbanization. Precise detection of heavy metals in soil offers valid reference for prevention and recovery of heavy metals in the field. In this research, Laser induced breakdown spectroscopy (LIBS) and chemometrics methods were employed to conduct quantitative analysis of heavy metals Pb and Cd in soil. Based on the pollution extent, soil samples with 15 concentration gradients of Pb and Cd were manually made up. Then, the LIBS emission lines of all soil samples were collected firstly. In order to eliminate errors and noise of spectral data, preprocessing methods called removal of abnormal data and normalization were used. Then, characteristic lines and spectral regions of Pb and Cd were determined based on our LIBS spectra and Atomic Spectra Database (ASD) of National Institute of Standards and Technology (NIST). Quantity regression models based on multiple linear regression (MLR), partial least squares regression (PLSR), least squares support vector machine (LS-SVM) and back propagation-artificial neural network (BP-ANN) were set up and their results were compared. As a result, models based on non-linear methods (LS-SVM and BP-ANN) offered a promising results than the linear methods of MLR and PLSR. The probable reason was that non-linear methods had an advantage to deal with matrix effects of soil automatically. The results indicated that LIBS coupled with multiple chemometrics methods provided a brand-new analysis approach for heavy metals accurate detection in soil and it could be considered as an effective theoretical foundation of making protection and recovery decision for soil contaminated by heavy metals.