Mirror neuron system dysfunction may underlie a self–other matching impairment, which has previously been suggested to account for autism. Embodied Cognition Theory, which proposes that action provides a foundation for cognition has lent further credence to these ideas. The hypotheses of a self–other matching deficit and impaired mirror neuron function in autism have now been well supported by studies employing a range of methodologies. However, underlying mechanisms require further exploration to explain how mirror neurons may be involved in attentional and mentalizing processes. Impairments in self–other matching and mirror neuron function are not necessarily inextricably linked and it seems possible that different sub-populations of mirror neurons, located in several regions, contribute differentially to social cognitive functions. It is hypothesized that mirror neuron coding for action–direction may be required for developing attentional sensitivity to self-directed actions, and consequently for person-oriented, stimulus-driven attention. Mirror neuron networks may vary for different types of social learning such as “automatic” imitation and imitation learning. Imitation learning may be more reliant on self–other comparison processes (based on mirror neurons) that identify differences as well as similarities between actions. Differential connectivity with the amygdala–orbitofrontal system may also be important. This could have implications for developing “theory of mind,” with intentional self–other comparison being relevant to meta-representational abilities, and “automatic” imitation being more relevant to empathy. While it seems clear that autism is associated with impaired development of embodied aspects of cognition, the ways that mirror neurons contribute to these brain–behavior links are likely to be complex.