Most leading research in basic and clinical neuroscience has been carried out by functional magnetic resonance imaging (fMRI), which detects the blood oxygenation level dependent signals associated with neural activities. Among new fMRI applications, brain decoding is an emerging research area, which infers mental states from fMRI signals. Brain decoding using fMRI includes classification, identification, and reconstruction of brain states. It is generally conducted using multi-voxel pattern analysis based on neuroscientific evidence that brain functions are mediated by distributed activation patterns. Brain decoding techniques have been successful in diverse applications such as the brain computer interface, patient monitoring, and neurofeedback. These techniques have expanded our understanding of how the brain encodes distinct information. In the current paper, we reviewed recent fMRI-based brain decoding techniques and applications. We also discussed the potential implications of brain decoding in neuroscience.