Laser-induced breakdown plasma of Pb concentration in the national standard soil samples is obtained by using 1064 nm Nd∶YAG laser,and plasma spectra of the samples are collected with fiber optic spectrometer.Quantitative calibration models for determination of Pb concentration in soil sample are established by partial least square (PLS) regression.The information is extracted and analyzed by applying a variety of spectral pretreatment methods.The effects of spectral pretreatment methods on the predicted ability of PLS models are discussed.The results show that models based on 9 spectral smoothing and multiplicative scatter correction have better quality.Correlation coefficients (R) of prediction for calibration samples and prediction samples are 0.9981 and 0.9948 respectively,root mean square error of cross validation (RMSECV) and root mean square error of prediction are 12 and 17 respectively.The deviation of Pb concentration between predicted value and actual measurement is within 10.5%.Therefore the feasibility of this PLS model for instant detection of heavy metals in soil is verified.