OBJECTIVE This study investigated the accuracy of physical therapists' visual judgments about kinematic features of the upper-limb movements of people without neurological impairments and people with neurological impairments following a cerebrovascular accident (CVA). METHODS Ten experienced physical therapists served as observers. Eleven people with a primary diagnosis of cortical or subcortical CVA and 4 older individuals without neurological impairments participated as "performers." METHODS The performers were videotaped as they completed a transport task. Three videotapes were edited to form three physical scales of peak movement speed, jerkiness, and hand path indirectness. On two occasions, therapists viewed these videotapes and made judgements about each performance on visual analog scales. Therapists' visual judgments were then compared with criterion measures determined by three-dimensional instrumented analysis. RESULTS The accuracy of the therapists' judgments was investigated using regression methods. Therapists were able to make moderately to highly accurate judgments of movement speed (r > or = .87), jerkiness (r > or = .78), and hand path indirectness (r > or = .68). These judgements remained highly stable over time (r > or = .82). Differences in therapists' judgement models, evident from slope and intercept variability in the regression models, were reflected in lower intertherapist agreement (intraclass correlation coefficients = .65-.85). CONCLUSIONS Experienced physical therapists accurately and reliably judged kinematic aspects of performance using observational assessment. Observational kinematic assessment warrants further systematic investigation.