To assess the association between dementia and risk of hospital readmission and to evaluate whether the effect of dementia on hospital readmission varies according to primary diagnosis. Retrospective cohort study. Nationwide discharge database of acute care hospitals in Japan. Individuals aged 65 and older diagnosed with one of the 30 most common diagnoses and discharged from 987 hospitals between April 2014 and September 2015 (N = 1,834,378). The primary outcome was unplanned hospital readmission within 30 days. Poisson generalized estimating equation models were fitted to assess the risks of readmission for individuals with and without dementia, using primary diagnosis as a possible effect modifier and clinical factors as potential confounders. The overall prevalence of dementia was 14.7% and varied according to primary diagnosis, ranging from 3.0% in individuals with prostate cancer to 69.4% in those with aspiration pneumonia. Overall, individuals with dementia had a higher risk of hospital readmission (8.3%) than those without (4.1%) (adjusted risk ratio (aRR])=1.46, 95% confidence interval (CI)=1.44-1.49), although diagnostic category substantially modified the relationship between dementia and hospital readmission. For hip fracture, dementia was associated with greater risk of hospital readmission (adjusted risk 11.5% vs 7.9%; aRR=1.46; 95% CI=1.28-1.68); this risk was attenuated for cholecystitis (adjusted risk 12.8% vs 12.4%; aRR=1.03; 95% CI=0.90-1.18). Risk of hospital readmission associated with dementia varied according to primary diagnosis. Healthcare providers could enforce interventions to minimize readmission by focusing on comorbid conditions in individuals with dementia and specific primary diagnoses that increase their risk of readmission.