The potential of mid-infrared (MIR) spectroscopy in combination with partial least-squares (PLS) regression was investigated to predict the soil sorption (distribution) coefficient (K(d)) of a nonionic pesticide (diuron). A calibration set of 101 surface soils collected from South Australia was utilized for reference sorption data and MIR spectra. Principal component analysis (PCA) was performed on the spectra to detect spectral outliers. The MIR-PLS model was developed and validated by dividing the initial data set into four validation sets. The model resulted in a coefficient of determination (R2) of 0.69, a standard error (SE) of 5.57, and a residual predictive deviation (RPD) of 1.63. The normalized sorption coefficient for the organic compound (K(oc)) approach, on the other hand, resulted in R2, SE, and RPD values of 0.42, 7.26, and 1.25, respectively. However, the significant statistical difference between the two models was mainly due to two outliers detected via PCA. Apart from spectral outliers, the performance of the two models was essentially similar for the rest of the calibration set. Outlier detection by the MIR-PLS model may gainfully be employed as a tool for improving prediction of K(d). The MIR-based model can provide a direct estimation of K(d) values based on the integrated properties of organic and mineral matter reflected in the infrared spectra.