Nutritional evaluation may predict clinical outcomes, such as hospital length of stay (LOS). We aimed to assess the value of nutritional risk and status methods, and to test standard anthropometry percentiles v. the 50th percentile threshold in predicting LOS, and to determine nutritional status changes during hospitalisation and their relation with LOS. In this longitudinal prospective study, 298 surgical patients were evaluated at admission and discharge. At admission, nutritional risk was assessed by Nutritional Risk Screening-2002 (NRS-2002), Malnutrition Universal Screening Tool (MUST) and nutritional status by Subjective Global Assessment (SGA), involuntary % weight loss in the previous 6 months and anthropometric parameters; % weight loss and anthropometry were reassessed at discharge. At admission, risk/undernutrition results by NRS-2002 (P< 0·001), MUST (P< 0·001), % weight loss (P< 0·001) and SGA (P< 0·001) were predictive of longer LOS. A mid-arm circumference (MAC) or a mid-arm muscle circumference (MAMA) under the 15th and the 50th percentile, which was considered indicative of undernutrition, did predict longer LOS (P< 0·001); conversely, there was no association between depleted triceps skinfold (TSF) and longer LOS. In-hospital, there was a high prevalence of weight, muscle and fat losses, associated with longer LOS. At discharge, patients with a simultaneous negative variation in TSF+MAC+MAMA (n 158, 53 %) had longer LOS than patients with a TSF+MAC+MAMA positive variation (11 (8–15) v. 8 (7–12) d, P< 0·001). We concluded that at risk or undernutrition evaluated by all methods, except TSF and BMI, predicted a longer LOS. Moreover, MAC and MAMA measurements and their classification according to the 50th percentile threshold seem reliable undernutrition indicators.