Objectives Various informant‐based questionnaires are used in clinical practice to screen for pre‐stroke cognitive problems. However, there is no guidance on which tool should be preferred. We compared the validity of the two most commonly used informant‐based tools. Methods We recruited consecutively admitted stroke patients. Patients' informants completed the Informant Questionnaire for Cognitive Decline in the Elderly Short Form (IQCODE‐SF, 16‐item) and Ascertain Dementia 8 (AD8). We assessed construct validity (accuracy) against a semi‐structured clinical interview for dementia or mild cognitive impairment (MCI), describing test accuracy metrics and comparing area under ROC curves (AUROC). We described criterion validity by evaluating associations between test scores and neuroimaging markers of dementia and overall ‘brain frailty’. Finally, we described prognostic validity comparing ROC curves for 18‐month clinical outcomes of dementia, death, stroke, and disability. Results One‐hundred‐thirty‐seven patient‐informant dyads were recruited. At usual clinical cut‐points, the IQCODE‐SF had comparable sensitivity to the AD8 (both = 92%) for pre‐stroke dementia, but superior specificity (IQCODE‐SF: 82% vs. AD8: 58%). Youden index suggested that the optimal AD8 threshold for diagnosis of dementia is ≥4. The IQCODE‐SF demonstrated stronger associations with markers of generalised and medial‐temporal lobe atrophy, neurovascular disease, and overall brain frailty. IQCODE‐SF also demonstrated greater accuracy for predicting future dementia (IQCODE‐SF AUROC = 0.903, 95% CI = 0.798–1.00; AD8 AUROC = 0.821, 95% CI = 0.664–0.977). Conclusions Both IQCODE‐SF and AD8 are valid measures of pre‐stroke dementia. Higher cut points for AD8 may improve performance in the acute stroke setting. Based on consistent superiority across a range of validity analyses, IQCODE‐SF may be preferable to AD8 for pre‐stroke dementia screening.