Soil lead (Pb) contamination by anthropogenic and industrial activities is a problem of global concern. In this research the possibility to adapt mid infrared-diffuse reflectance infrared Fourier transform spectroscopy (MIR-DRIFTS) approach for the quantitative estimation of Pb in polluted soils was explored. One hundred soil samples were collected from an urban landfill agricultural site and scanned by MIR-DRIFTS. The raw reflectance spectra were preprocessed using four spectral transformations for predicting soil Pb contamination using three multivariate algorithms. Partial least squares regression using Savitzky-Golay (SG) first derivative spectra (RPD = 3.05) outperformed principal component regression models. The artificial neural networks-SG model using an independent validation set produced satisfactory generalization capability (RPD = 2.01). Thus, the combination of MIR-DRIFTS and multivariate models can reduce chemical analysis frequency for soil pollution monitoring, substantially reducing labor and analytical cost.