Direct observations of the living human brain in-vivo in-situ are are difficult to obtain, and consequently, a lot of our understanding of the cerebral metabolism under different conditions is inevitably indirect and model-based. A complex model of the cerebral metabolism depends on a large number of unknown parameters; on the other hand, reducing the model often impoverishes it, making it unsuitable for out-of-sample predictions. For this reason, it is desirable to try to quantify the uncertainties of the metabolic models, to exclude over interpretations based on a choice of poorly validated model parameters. In this talk we present some ideas on how uncertainty in brain metabolic states can be quantified, and suggest how to account for model and data based uncertainties in model predictions.