Objectives The Cohen‐Mansfield Agitation Inventory (CMAI; (Cohen‐Mansfield and Kerin, 1986)) is a well‐known tool for assessing agitated behaviours in people with dementia who reside in long‐term care. No studies have evaluated the psychometric qualities and factor structure of the CMAI in acute general hospitals, a setting where people with demand may become agitated. Method Longitudinal study investigating pain, agitation and behavioural problems in 230 people with dementia admitted to acute general hospitals in 2011–2012. Cohen‐Mansfield Agitation Inventory was completed as part of a battery of assessments including PAINAD to measure pain. Results A nine‐item two‐factor model of aggressive and nonaggressive behaviours proved to be the best‐fitting measurement model in this sample, (χ2 = 96.3, df = 26, p<0.001; BIC [Bayesian Information Criterion] = 4593.06, CFI [Comparative Fit Index] = 0.884, TLI [Tucker Lewis Index] = 0.839, RMSEA [Root Mean Square Error of the Approximation] = 0.108). Although similar to the original factor structure, the new model resulted in the elimination of item 13 (screaming). Validity was confirmed with the shortened CMAI showing similar associations with pain as the original version of the CMAI, in particular the link between aggressive behaviours and pain. Conclusion The factor structure of the CMAI was broadly consistent with the original solution although a large number of items were removed. Scales reflecting physical and verbal aggression were combined to form an Aggressive factor, and physical and verbal nonaggressive behaviours were combined to form the Nonaggressive factor. A shorter, more concise version of the CMAI was developed for use in acute general hospital settings. Copyright © 2017 John Wiley & Sons, Ltd.