In this paper, we present a method for classifying functional near-infrared spectroscopy (fNIRS) data using wavelets and support vector machine (SVM). fNIRS data is acquired by ETG-4000 during speed and force imagination. Probes location is around C3 and C4 in 10–20 international system. After preprocessing the data using NIRS-SPM, we decompose it with ‘db5’ wavelet for 9 levels to do a multiresolution analysis (MRA). Then, the approximation and detail signal at every level are used for SVM classification using libSVM toolbox. The results show that frequency band between 0.02 and 0.08Hz is important for classification, especially frequency band between 0.02 and 0.04Hz. This finding is useful for building an fNIRS-based brain computer interface (BCI) system.