Soil salinization is a common desertification process, especially in arid lands. Hyperspectral remote sensing of salinized soil is favored for its advantages of being efficient and inexpensive. However, soil moisture often jointly has a great influence on the soil reflectance spectra under field conditions. It is a challenge to establish a model to eliminate the effect of soil moisture and quantitatively estimate the salinity contents of slightly and moderately salt-affected soil. A controlled laboratory experiment was conducted by way of continuously monitoring changes of soil moisture and salt content, which was mainly focused on the slightly and moderately salt-affected soil. We investigated the external parameter orthogonalization (EPO) method to remove the effect of soil moisture (4 to 36% in weight base) by preprocessing soil spectral reflectance and establishing the partial least squares regression after EPO preprocessing model (EPO-PLS) to predict soil salt content. Through comparing PLS with EPO-PLS model, and ratio of prediction to deviation rose from 0.604 and 1.063, respectively, to 0.874 and 2.865 for validation data. Root mean square error and bias were, respectively, reduced from 1.163 and to 0.718 and . The performance of the model after EPO algorithm preprocessing was improved significantly.