There is a demand for a new neurorehabilitation modality with a brain-computer interface for stroke patients with insufficient or no remaining hand motor function. We previously developed a robotic hand rehabilitation system triggered by multichannel near-infrared spectroscopy (NIRS) to address this demand. In a preliminary prototype system, a robotic hand orthosis, providing one degree-of-freedom motion for a hand's closing and opening, is triggered by a wireless command from a NIRS system, capturing a subject's motor cortex activation. To examine the feasibility of the prototype, we conducted a preliminary test involving six neurologically intact participants. The test comprised a series of evaluations for two aspects of neurorehabilitation training in a real-time manner: classification accuracy and execution time. The effects of classification-related factors, namely the algorithm, signal type, and number of NIRS channels, were investigated. In the comparison of algorithms, linear discrimination analysis performed better than the support vector machine in terms of both accuracy and training time. The oxyhemoglobin versus deoxyhemoglobin comparison revealed that the two concentrations almost equally contribute to the hand motion estimation. The relationship between the number of NIRS channels and accuracy indicated that a certain number of channels are needed and suggested a need for a method of selecting informative channels. The computation time of 5.84 ms was acceptable for our purpose. Overall, the preliminary prototype showed sufficient feasibility for further development and clinical testing with stroke patients.