Fast classification of soil with different texture is essential for site-specific application of different inputs into farmland. Total 178 soil samples with five textures were collected from Silsoe Farm, Cranfield University, England. Using a Vis/NIR spectrophotometer (LabSpec2500, ASD), spectra of soil samples were scanned for the study. Amongst various pre-processing methods, smoothing with moving average(MA), standard normal variation(SNV) and 1st derivatives were mainly investigated. PCA was applied to evaluate the discriminative capacity of the pre-processing methods on soil spectra. The score plot of PCs shows that 1st derivative with variable smoothing points can help classify soil samples much more effectively than others. With the increase of smoothing points, the cumulative variance of first few PCs in PCA tends to increase while the discriminative capability based on these PCs becomes worse. A trade-off between cumulative variance and discriminative capability should be concerned. In the study, an appropriate range of smoothing points in the 1st derivative is 7-21.