Fast qualification of soil organic matter (SOM) is important to crop production and evaluation of soil quality. Fourier transform infrared photoacoustic spectroscopy (FTIR-PAS) makes it feasible to quantify soil organic matter content in a rapid way. In this study, FTIR-PAS was applied to measure SOM in the soils collected from paddy fields in Lishui District of Jiangsu Province. Support vector machine (SVM) was utilized to build calibration models. Principal component analysis (PCA), partial least squares (PLS) and independent component analysis (ICA) were performed separately to extract principal components (PCs), latent variables of PLS (LVPLS) and independent components (ICs) from the soil spectra as input of support vector machine (SVM). Hence, three SVM calibration models were built up. Meanwhile, PLS was also used to form a calibration model as control. Results show that the ICs-based SVM model performed best in prediction of SOM, with correlation coefficient (R2), root-mean-square error (RMSEP) and ratio of performance to standard deviation (RPD) being 0.808, 0.575 and 2.28 respectively. Furthermore, F-test demonstrates that this model was significantly superior over the PCs-based SVM model, but was quite similar to the LVPLS-based SVM model and the classic PLS-based SVM model. Besides, no significant difference was observed between the predictions using the calibration models and the determination using the chemical method, as was demonstrated by t-test. It can, therefore, be concluded that the technology of infrared photoacoustic spectroscopy can be used as a new means for rapid determination of soil organic matter content.