Theory of mind (ToM) is assumed to be instrumental to social interactions, yet it is typically studied using non-interactive laboratory tasks. Standard measures are thus limited in their ability to characterize the cognitive and neural substrates of ToM in naturalistic social interactions, as well as the mechanisms explaining social-interactive difficulties in autism spectrum disorder (ASD). Across three studies, this dissertation aimed to highlight and bridge the disconnect between the study of ToM and its real-world implementation. Study 1 assessed the relative importance of a range of social-cognitive, social-perceptual, and social-affective constructs in explaining variance in the social symptoms of ASD. Three standard, non-interactive ToM measures together explained only 6% of the variance in social symptoms, reinforcing the need for interactive approaches to studying ToM. Study 2 applied such an approach using a socially interactive neuroimaging paradigm to measure brain activation associated with both ToM and social interaction. In typically developing children aged 8-12, interacting with a peer, even in the absence of explicit ToM demands, engaged many of the same regions as did non-interactive ToM reasoning, consistent with the idea that social interaction elicits spontaneous ToM-related processes. Study 3 also investigated ToM in social interaction, this time at the behavioral level, by introducing a novel observational coding system that measures the use of (or failure to use) ToM in naturalistic conversation. Among dyads of typically developing and autistic children and adolescents, conversational ToM (cToM) did not predict interaction success. However, the cToM Negative subscale—capturing ToM-related violations of conversational norms—was negatively associated with two forms of non-interactive ToM: 1) recognizing complex emotions from facial expressions, and 2) spontaneously attributing mental states when describing abstract social animations. Furthermore, exploratory analyses revealed associations between cToM and brain activation during the socially interactive neuroimaging task used in Study 2. Findings across the three studies highlight the multifaceted nature of the ToM construct, the value of socially interactive approaches to studying ToM, and the importance of considering ToM alongside other social-cognitive and affective processes when investigating social interaction.