The present study determined the ability of the Mini Nutritional Assessment (MNA) to predict care need in older people. We analysed the datasets of the Taiwan Longitudinal Study on Aging. The 1999 survey containing the MNA items served as the baseline and the 2003 survey served as the endpoint. Of the 4440 participants, 2890 were aged  ≥ 65 years and served as subjects in the present study. After excluding 150 subjects having incomplete data, 2740 were rated for nutritional status with the normalised long-form (LF) and short-form (SF) MNA-Taiwan version 1 (T1) and version 2 (T2) and evaluated with logistic regression analysis for cross-sectional associations of the rated nutritional status with care need, controlled for age, sex, education level, living arrangement and physical activity. Receiver operating characteristic curves were generated for evaluating the ability of the MNA to predict care need. After further excluding 250 subjects who had care need at baseline and seventy-six who were lost to follow-up, 2414 were evaluated for the ability of the MNA to predict subsequent care need with logistic regression analysis. The results demonstrated that all the MNA predicted concurrent and subsequent care need well. The OR for needing subsequent care in the ‘at-risk’ and ‘malnourished’ groups were, respectively, 2·04 and 3·33 for the MNA-T1-LF, 2·10 and 5·35 for the MNA-T2-LF, 1·49 and 2·48 MNA-T1-SF, and 1·80 and 3·44 for the MNA-T2-SF (all P< 0·05), and the respective Nagelkerke R2 values were 0·190, 0·191, 0·184 and 0·192. In conclusion, all the four MNA have the ability to predict future care need, including the MNA-T2-SF, which appears to have great potential for practical applicability.