Laser-induced breakdown spectroscopy is used to measure chromium concentration in soil samples. A comparison is carried out between the calibration curve method and two chemometrics techniques: partial least-squares regression and neural networks. The three quantitative techniques are evaluated in terms of prediction accuracy, prediction precision, and limit of detection. The influence of several parameters specific to each method is studied in detail, as well as the effect of different pretreatments of the spectra. Neural networks are shown to correctly model nonlinear effects due to self-absorption in the plasma and to provide the best results. Subsequently, principal components analysis is used for classifying spectra from two different soils. Then simultaneous prediction of chromium concentration in the two matrixes is successfully performed through partial least-squares regression and neural networks.