Many characteristics typical of autism, a neurodevelopmental condition characterised by socio-communicative impairments, are most evident during social interaction. Accordingly, procedures such as the Autism Diagnosis Observation Schedule (ADOS) are interactive and intended to elicit interactional impairments: a diagnosis of autism is given if interactional difficulties are attributed as a persistent quality of the individual undergoing diagnosis. This task is difficult, first, because behaviours can be interpreted in various ways and, second, because conversation breakdown may indicate a disengagement with, or resistance to, a line of conversation. Drawing upon conversation analysis, we examine seven ADOS diagnosis sessions and ask how diagnosticians distinguish between interactional resistance as, on the one hand, a diagnostic indicator and, on the other, as a reasonable choice from a range of possible responses. We find evidence of various forms of resistance during ADOS sessions, but it is a resistance to a line of conversational action that is often determined to be indicative of autism. However, and as we show, this attribution of resistance can be ambiguous. We conclude by arguing for reflexive practice during any diagnosis where talk is the problem, and for a commitment to acknowledge the potential impact of diagnostic procedures themselves upon results.