Functional near-infrared spectroscopy (fNIRS) is a promising technique for monitoring brain activity. However, it is sensitive to motion artifacts. Many methods have been developed for motion correction, such as spline interpolation, wavelet filtering, and kurtosis-based wavelet filtering. We propose a motion correction method based on empirical mode decomposition (EMD), which is applied to segments of data identified as having motion artifacts. The EMD method is adaptive, data-driven, and well suited for nonstationary data. To test the performance of the proposed EMD method and to compare it with other motion correction methods, we used simulated hemodynamic responses added to real resting-state fNIRS data. The EMD method reduced mean squared error in 79% of channels and increased signal-to-noise ratio in 78% of channels. Moreover, it produced the highest Pearson’s correlation coefficient between the recovered signal and the original signal, significantly better than the comparison methods (p<0.01, paired t-test). These results indicate that the proposed EMD method is a first choice method for motion artifact correction in fNIRS.