Aim Impaired attention is a core diagnostic feature for delirium. The present study examined the discriminating properties for patients with delirium versus those with dementia and/or no neurocognitive disorder of four objective tests of attention: digit span, vigilance “A” test, serial 7s subtraction and months of the year backwards together with global clinical subjective rating of attention. Methods This as a prospective study of older patients admitted consecutively in a general hospital. Participants were assessed using the Confusion Assessment Method, Delirium Rating Scale–98 Revised and Montreal Cognitive Assessment scales, and months of the year backwards. Pre‐existing dementia was diagnosed according to the Diagnostic and Statistical Manual of Mental Disorders fourth edition criteria. Results The sample consisted of 200 participants (mean age 81.1 ± 6.5 years; 50% women; pre‐existing cognitive impairment in 126 [63%]). A total of 34 (17%) were identified with delirium (Confusion Assessment Method +). The five approaches to assessing attention had statistically significant correlations (P < 0.05). Discriminant analysis showed that clinical subjective rating of attention in conjunction with the months of the year backwards had the best discriminatory ability to identify Confusion Assessment Method‐defined delirium, and to discriminate patients with delirium from those with dementia and/or normal cognition. Both of these approaches had high sensitivity, but modest specificity. Conclusion Objective tests are useful for prediction of non‐delirium, but lack specificity for a delirium diagnosis. Global attentional deficits were more indicative of delirium than deficits of specific domains of attention. Geriatr Gerontol Int 2016; 16: 1028–1035.