Introduction: Malnutrition is a frequent condition in hospitalized patients, and it is related to worse outcomes. The early diagnosis of malnutrition in hospitalized patients allows clinicians to target appropriate nutrition interventions to reduce complications. For this purpose, accurate tools should be adopted instead of body mass index (BMI). Then, we aimed to confirm the inaccuracy of BMI for malnutrition diagnosis in hospitalized patients, considering subjective global assessment (SGA) as the reference method. Methods: We conducted a cross-sectional study with secondary data from two longitudinal studies and analyzed data of patients aged > 18 years included in two cohorts conducted in two South-Brazilian hospitals. They were evaluated in the first 72 hours after hospital admission by a trained research team that diagnosed malnutrition by BMI and SGA. Analysis of accuracy and concordance were stratified by age, sex and ethnicity. Results: A total of 1,348 patients were included in the analysis (54.5±15.2 years, 51.9% females, and 82.4% of white self-reported ethnicity). The BMI identified malnutrition in 8.7% (n = 117) of the total sample, while the SGA diagnosed malnutrition in 32.4% (n = 437) of patients. BMI accuracy was unsatisfactory (AUC ROC curve=0.541; CI95% 0.508-0.574); and sensitivity (55.6%), specificity (67.7%), and positive predictive value (14.1%) were low, while negative predictive value (94.1%) was high. The Kappa coefficient showed a poor concordance between reduced BMI and SGA for malnutrition diagnosis (k = 0.100). The same pattern of results was observed in analysis stratified by age, sex and ethnicity. Conclusions: BMI was not accurate for malnutrition diagnosis in a large sample of patients, regardless of age, sex and ethnicity and these results confirm that BMI is inappropriate for nutritional assessment in clinical practice