BACKGROUND: Pain is common in people with dementia in general hospitals. This can be difficult to identify. OBJECTIVES: To evaluate the psychometric properties of PainChek electronic pain assessment tool. DESIGN: Cross-sectional psychometric study. SETTING: Six medical care of older people wards from two general hospitals in greater London, UK. SUBJECTS: 63 people with clinical diagnosis of dementia: mean 84 years (SD 6.7), 59% female, 69% living in their own homes, 64% white British, 77% moderate/severe dementia. METHOD: Psychometric evaluation of PainChek, a point-of-care electronic pain assessment tool combining artificial intelligence, facial analysis and smartphone technology. From a total of 216 assessments, we tested PainChek's inter-rater reliability (IRR) (Cohen's kappa), internal consistency (Cronbach's alpha) and concurrent validity (Pearson's coefficient) between PainChek and Pain Assessment in Advanced Dementia (PAINAD) scores at rest and post-movement [95% confidence interval (95% CI) where appropriate]. We assessed convergent validity with Symptom Management-End of Life in Dementia scale (SM-EOLD) (Pearson's coefficient) and discriminant validity (rest vs post-movement). RESULTS: IRR was 0.714 (95% CI 0.562 to 0.81) (rest) and 0.817 (95% CI 0.692 to 0.894) (post-movement). Internal consistency was 0.755 (rest) and 0.833 (post-movement). Concurrent validity with PAINAD was 0.528 (95% CI 0.317 to 0.690) (rest) and 0.787 (0.604 to 0.891) (post-movement). Convergent validity with SM-EOLD was -0.555 (95% CI -0.726 to -0.318) (rest) and -0.5644 (95% CI -0.733 to -0.331) (post-movement). Discriminant validity was significant. CONCLUSIONS: PainChek is a valid and reliable pain assessment tool for people with dementia in general hospitals. Further consideration will be needed for implementation into this setting.