Field measurement using NIR spectroscopy has become very popular for measuring soil properties. However NIR reflectance is quite sensitive to external environmental conditions, such as temperature, and soil moisture in particular. In field measurement, the soil moisture content can be highly variable. It is a challenge to find a method for predicting the properties of soil samples in the field that have variable moisture content. This paper attempts to develop a novel algorithm to remove the spectral effect of soil moisture for the calibration of soil carbon content. The algorithm projects all the soil spectra orthogonal to the space of variation. Here the unwanted variations of soil moisture can be effectively removed. We conducted experiments using soils at different moisture content, and the results show that it is feasible to remove the moisture effect from field spectra. This resulted in improved calibration of soil carbon content.