Routine delirium screening could improve delirium detection, but it remains unclear as to which screening tool is most suitable. We tested the diagnostic accuracy of the following screening methods (either individually or in combination) in the detection of delirium: MOTYB (months of the year backwards); SSF (Spatial Span Forwards); evidence of subjective or objective 'confusion'. We performed a cross-sectional study of general hospital adult inpatients in a large tertiary referral hospital. Screening tests were performed by junior medical trainees. Subsequently, two independent formal delirium assessments were performed: first, the Confusion Assessment Method (CAM) followed by the Delirium Rating Scale-Revised 98 (DRS-R98). DSM-IV (Diagnostic and Statistical Manual of Mental Disorders, fourth edition) criteria were used to assign delirium diagnosis. Sensitivity and specificity ratios with 95% CIs were calculated for each screening method. 265 patients were included. The most precise screening method overall was achieved by simultaneously performing MOTYB and assessing for subjective/objective confusion (sensitivity 93.8%, 95% CI 82.8 to 98.6; specificity 84.7%, 95% CI 79.2 to 89.2). In older patients, MOTYB alone was most accurate, whereas in younger patients, a simultaneous combination of SSF (cut-off 4) with either MOTYB or assessment of subjective/objective confusion was best. In every case, addition of the CAM as a second-line screening step to improve specificity resulted in considerable loss in sensitivity. Our results suggest that simple attention tests may be useful in delirium screening. MOTYB used alone was the most accurate screening test in older people.