Background Malnutrition may worsen clinical outcomes in stroke patients. Few malnutrition screening tools have been validated in the rehabilitation setting. The present study aimed to assess the concurrent and predictive validity of two malnutrition screening tools. Methods We retrospectively collected scores for the Mini Nutritional Assessment Short‐Form (MNA‐SF) and the Geriatric Nutritional Risk Index (GNRI) in consecutive stroke patients aged ≥65 years in a rehabilitation hospital. Concurrent validity was confirmed against the European Society for Clinical Nutrition and Metabolism diagnostic criteria for malnutrition (ESPEN‐DCM). Malnutrition risk within the ESPEN‐DCM process was assessed using the Malnutrition Universal Screening Tool. Cut‐off values with maximum Youden index, and with sensitivity (Se) >90% and specificity (Sp) >50%, were defined as appropriate for identification and screening of malnutrition, respectively. The Functional Independence Measure and discharge destination were used to explore predictive validity. Results Overall, 420 patients were analysed. Of these, we included 125 patients in the malnutrition group and 295 in the non‐malnutrition group based on the ESPEN‐DCM. Cut‐off values for the identification and screening of malnutrition were 5 (Se: 0.78; Sp: 0.85) and 7 (Se: 0.96; Sp: 0.57) for the MNA‐SF; 92 (Se: 0.74; Sp: 0.84) and 98 (Se: 0.93; Sp: 0.50) for the GNRI, respectively. The GNRI predicted discharge to acute care hospital, whereas the MNA‐SF did not predict all outcome measures. Conclusions The MNA‐SF and the GNRI have a fair concurrent validity in stroke patients, although lower cut‐off values than currently used were required for the MNA‐SF. The GNRI exhibits good predictive validity for discharge destination.