Direct observation of peer relating is potentially a sensitive and ecologically valid measure of child social functioning, but there has been a lack of standardised methods. The Manchester Inventory for Playground Observation (MIPO) was developed as a practical yet rigorous assessment of this kind for 5–11 year olds. We report on the initial reliability and validity of the MIPO and its ability to distinguish social impairments within different psychopathologies. We observed 144 clinically referred children aged 5;00–11;11 (mean 8.8) years with Externalising (n=44), Internalising (n=19), Autism Spectrum Disorders (n=39) or Specific Language Impairment (n=42), and 44 class-controls, in naturalistic playground interaction. Observers, blind to clinical diagnosis, completed the MIPO and the teacher checklist from the Social Skills Rating System (SSRS). MIPO items showed high internal consistency (alpha=.924; all ‘alpha if item deleted’ values>.91), inter-observer reliability (mean κw=.77) and test–retest stability (over 2 weeks; mean κw=.58). MIPO totals showed convergence with SSRS (n=68, rs=.78, p<.01) and excellent discrimination between case and control (sensitivity=0.75 and specificity=0.88, AUC=.897). Externalising, Autistic Spectrum and Language Impaired groups showed distinct profiles of MIPO impairment consistent with theory:Internalising disorders less so. 65.3% of clinical cases were classified accurately for primary diagnosis. The MIPO shows reliability and validity as a measure of children's social functioning relevant in developmental research and as a clinical tool to aid differential diagnosis and intervention planning.