The ability to flexibly modulate our behaviors in social contexts and to successfully interact with other persons is a fundamental, but pivotal, requirement for human survival. Deficit in this ability is naturally associated with socially isolating mental diseases, e.g., autism. Although ample studies have examined different aspects of autism, the diagnosis and treatment of individuals with ASD are still mainly based on behavioral observations and assessments. To investigate the neural bases of autism and to find the biomarkers for diagnosis/treatment, various neuroimaging techniques have been used in the fields of autism research. However, due to the technical limitations of brain-imaging techniques, which are not appropriate to measure brain activations in children with ASD in realistic social contexts. Most of studies have focused on high-functioning children with ASD over 3 years old in the resting state or in the simple tasks, involving only the passive observation of emotion or social relevant pictures or video clips. That is, the neuroimaging research on autism, a social deficits disorder, in non-social contexts. To fully understand the neural substrates underlying social deficits in individuals with ASD, it is necessary to examine their brain activations in natural social communications or interactions, and focus not only on intra-brain activation or connectivity but also on inter-brain connectivity considering the interacting members as a coupled unit. Functional near-infrared spectroscopy (fNIRS) is a relatively new, noninvasive brain-imaging technique for studying functional activation by measuring changes in the hemodynamic properties of the brain. The fNIRS has few physical constraints and is more tolerant to motion artifacts and electromagnetic noise, permitting simultaneous measurements of two or more persons’ brains in relatively natural social contexts. Thus, fNIRS is suitable for neuroimaging research on children, even on infant within the first year, which will contribute greatly for understanding of autism since its symptoms normally occur before 18 months. Especially, a recent new technique of hyperscanning allows simultaneously measurement of two or more individuals in real social interactions, providing an opportunity to explore the neural mechanisms of social deficits in children with ASD during interactive activities. In this paper, thus, we review studies using fNIRS on children with ASD and social interactions to show the potential importance of fNIRS to help to solve the puzzle of autism. Specifically, previous fNIRS studies on autism and interpersonal interactions suggest that the atypical intra-brain activation and connectivity in the prefrontal cortex (PFC) and the mirror neuron system (MNS) may play important roles in social deficits of the individuals with ASD. But it needs careful investigation on issue of the causal relationships between the roles of behavioral deficits in children with ASD and the atypical functional activation and connectivity in the PFC and the MNS. This could be addressed by studying the children with ASD at different ages from the first year of their life using both cross-sectional and longitudinal designs. Furthermore, it will be necessary to examine inter-brain connectivity across individuals with ASD and others in real social interactions. In this case, using fNIRS and hyperscanning techniques, we could firstly "look at" autistic children's brains when they are engaging in social activities with their parents, friends or teachers, which in turn allows understanding of their brains from new perspectives.