Uncertainty in basic data and errors in a model's structure constitute the two causes of prediction error for a deterministic system. Quantification of uncertainty in all basic data is a complicated job, whereas structural errors can only be studied under the assumption that some given, presumably complex, model is structurally correct. The two sources of prediction error can be analyzed separately, but their joint study may be of help in selecting a suitably simplified model structure for a specific application. With given uncertainties in basic data, complexification of a model beyond some degree becomes futile and may even increase prediction error. By the imperfections of the analysis of input uncertainties and structural errors, model validation remains indispensable, preferably through empirical assessment of error in the most relevant predictions.