Intentional harms are typically judged to be less forgivable than accidental harms. This difference depends on mental state reasoning (i.e., reasoning about beliefs and intentions), supported by a group of brain regions, the ‘theory of mind’ network. Prior research has found that (i) interfering with activity in this network can shift moral judgments away from reliance on mental state information, and (ii) high-functioning individuals with Autism Spectrum Disorder (ASD) rely significantly less on mental state information to make moral judgments than matched neurotypical (NT) participants. Across three experiments, we find using multi-voxel pattern analysis (MVPA) that, in NT adults, (i) one key region of the ToM network, the RTPJ, shows reliable and distinct spatial patterns of responses across voxels for intentional versus accidental harms, and (ii) individual differences in this neural pattern predict individual differences in moral judgment. By contrast, (iii) in ASD adults, the difference between intentional and accidental harms is not encoded in the voxel- wise pattern in the RTPJ or any other region, and (iv) higher symptom severity scores are predictive of diminished pattern discriminability. We conclude that MVPA can detect features of mental state representations and that these features are behaviorally and clinically relevant.