Introduction: Studies of delirium after acute stroke focus on stroke units (SUs). A protective effect of SUs against delirium has been suggested. We hypothesized that selection bias against medically complex patients accounts for this apparent effect. Methods: An observational cohort of acute ischemic stroke patients was screened for post-stroke delirium. Delirium was diagnosed using the Confusion Assessment Method (CAM). Key patient variables were prospectively recorded including initial NIHSS score and medical complications. Univariate associations with delirium were identified and a logistic regression model was developed for the entire cohort. Separate logistic regression models were also developed for non-stroke unit (NSU) and SU patients. The SU consisted of a specialized stroke ward, step-down stroke unit, and a neuroscience ICU. Results: Over 10 months 246 patients (56% male, mean 65 years, 29% in NSUs) met inclusion criteria. Delirium occurred in 30 (12%) patients and was less frequent in the SU (8.0% vs 22.5%, p=0.002). Frequency of CAM checks differed between NSU and SU (median 4.1 vs 3.7 per day; p=0.03). NSU patients had similar NIHSS scores as SU patients (median [interquartile range], 3 [1-8] vs 3 [1-5]; p=0.18) but more ICU admissions (48% vs 27%, p=0.001) and more infections (18% vs 9.7%, p=0.06). In the entire cohort, initial NIHSS (OR 1.07, 95% CI 1.02-1.13; p=0.006), cardioembolic stroke mechanism (OR 3.0, 95% CI 1.3-6.9; p=0.009) and SU care (OR 0.39, 95% CI 0.17-0.88; p=0.02) predicted delirium after correcting for covariates, including frequency of CAM checks. In the NSU model, age (OR 1.06, 95% CI 1.01-1.11; p=0.02) and infections (OR 6.8, 95% CI 1.5-30.2; p=0.01) were associated with delirium. Only cardioembolic stroke mechanism (OR 5.4, 95% CI 1.7-16.7; p=0.003) was associated with delirium in the SU model. Conclusion: Associations with delirium after acute ischemic stroke differ between NSU and SU patients. Stroke patients treated in NSUs are fundamentally different than SU patients. Given the potential for residual confounding, the lower incidence of delirium after stroke in SUs than NSUs may reflect selection bias rather than a specific effect of SU care.