Infrared partial least squares (PLS) analysis is shown to provide a simple, rapid chemometric technique for the simultaneous analysis of soil properties. The method is capable of extracting both qualitative and quantitative information from soil spectra. A number of the mineral and organic components which are responsible for certain soil properties have been identified and the prediction of these properties assessed. Diffuse reflectance infrared Fourier-transform (DRIFT) spectra of whole soils were recorded to form a large training data set. The spectral information from this set was compressed into a small number of subspectra (called weight loadings) which contained positive and negative peaks reflecting correlations between the soil mineral and organic components and corresponding analytical data. Positive peaks in the weight loadings corresponding to organic components including alkyl, carboxylic and amide species were highly correlated with OC and N. Likewise, smectite, kaolinite and gibbsite clay minerals, together with organic alkyl and carboxylic species, contributed either positively or negatively to pH, sum of cations and clay content. Positive peaks due to calcite were well resolved in the first carbonate weight loading. Quartz was identified as an 'interference' for all analyses, with a series of negative peaks in the weight loadings. Implications were that quartz exerted a strong spectral signal for the majority of soil spectra, although it was not directly related to particular analyses. The usual PLS method, in which there is assumed to be a linear relationship between the loading intensities and soil property values, was found to give nonlinear prediction regression. The nonlinearity was assumed to be due to the effects of nonlinear response of the DRIFT signal and to significant compositional variability between calibration samples with high and low analyte concentrations. An alternative strategy of using a locally linear PLS model was tested, where small subsets of the total span of analytical values were independently used for PLS analysis. This approach improved prediction linearity and precision improved significantly for most analyses. The PLS method was thus shown to provide a useful surrogate technique for the study of soils, with which the PLS analysis of a single spectrum could provide simultaneous qualitative and quantitative information on a number of widely different soil analyses.