Objective: To investigate the effect of hand action observation training, i.e., mirror neuron system (MNS) based training, on language function of aphasic patients after stroke. In addition, to reveal the tentative mechanism underlying this effect.Methods: Six aphasic patients after stroke, meeting the criteria, undergo 3 weeks' training protocol (30 min per day, 6 days per week). Among them, four patients accepted an ABA training design, i.e., they implemented Protocol A (hand action observation combined with repetition) in the first and third weeks and carried out Protocol B (static object observation combined with repetition) in the second week. Conversely, for the other two patients, BAB training design was adopted, i.e., patients took Protocol B in the first and third weeks and accepted Protocol A in the second week. Picture naming test, western aphasia battery (WAB) and Token Test were applied to evaluate the changes of language function before and after each week's training. Furthermore, two subjects (one aphasic patient and one healthy volunteer) attended a functional MRI (fMRI) experiment, by which we tried to reveal the mechanism underlying possible language function changes after training.Results: Compared with static object observation and repetition training (Protocol B), hand action observation and repetition training (Protocol A) effectively improved most aspects of the language function in all six patients, as demonstrated in the picture naming test, subtests of oral language and aphasia quotient (AQ) of WAB. In addition, the fMRI experiment showed that Protocol A induced more activations in the MNS of one patient and one healthy control when compared to Protocol B.Conclusion: The mirror neuron based therapy may facilitate the language recovery for aphasic patients and this, to some extent, provides a novel direction of rehabilitation for aphasia patients.