In order to determine whether surrogate markers predict clinical outcome, randomized controlled trials (RCT) of nutrition support v. no nutrition support that have reported at least one clinical outcome (mortality, infections, total complications, or duration of hospitalization) and at least one nutritional outcome (energy or protein intake, weight gain, N balance, albumin, prealbumin, transferrin, three anthropometric measures, skin testing, lymphocyte count) were assessed for concordance. If changes in nutritional markers predict clinical outcome, changes in both outcomes should go in the same direction. Concordance is defined as both outcomes changing in the same direction or both outcomes showing no difference. Discordance is defined as one outcome changing and the other not (partial) or both outcomes changing in opposite directions (complete). Ninety-nine RCT were identified, of which most were underpowered to see statistically significant changes, especially in clinical outcomes. Thus, the results were analysed only in relation to the direction of the respective changes in outcomes. Forty-eight comparisons (4×12) were made. The rates of concordance were ≤50% in forty-one of forty-eight comparisons; the rate was never >75%. A complete discordance rate of ≥25% was present in forty-three (≥50% in thirteen) of the forty-eight comparisons. The discordance was usually a result of the nutritional outcome being better than the clinical outcome. Changes in nutritional markers do not predict clinical outcomes. Before adopting any surrogate marker as an end point for a clinical trial, it has to be known that improving it will result in patient benefit.