Our aim was to investigate the long-term prognosis of delirium in the frailest elderly, and to clarify whether delirium is just a marker of the underlying severe disease. We used logistic regression analysis to determine the independent prognostic significance of delirium. A representative sample of 425 patients (≥70 years) in acute geriatric wards and nursing homes were assessed at baseline and followed up for 2 years. DSM-IV was used for classification. The prevalence of delirium at baseline was 24.9% (106/425). The prognosis of delirium was poor: mortality at 1 year was 34.9 vs. 21.6% in nondelirious subjects (p = 0.006), and at 2 years 58.5 vs. 42.6% (p = 0.005). Among home-dwelling people at baseline, 54.4% of the delirious vs. 27.9% of others were permanently institutionalized within 2 years (p < 0.001). In logistic regression analysis, delirium was an independent predictor for mortality at 1 year (OR 1.86, 95% CI 1.1–3.1), at 2 years (OR 1.76, 95% CI 1.1–2.8), and for permanent institutionalization (OR 2.45, 95% CI 1.2–4.9). Delirious patients with prior dementia tended to have a better prognosis than those without.