The parsimony principle is a useful criterion for reducing the non-uniqueness in classical system identification. However, if a uniquely selected model is used for prediction, the disregard of the uncertainty in model structure can lead to an underestimation of the uncertainty in model forecasts. This is particularly the case when processes become important during the prediction period that were insignificant during the identification period. If some knowledge of such processes is available, they should be included in the analysis. This requires an identification and forecasting technique that can use prior knowledge and can handle overparameterized, non-identifiable models. The Bayesian approach to statistical inference is such a technique. In this paper, the advantages and disadvantages of both the classical and the Bayesian methodology are discussed, and it is argued that from a methodical point of view, for poorly identifiable systems typical in ecological modelling, the Bayesian technique is the superior approach. Because of the huge computational requirements of the Bayesian technique a recommendation is given for an improved identification and forecasting procedure that, depending on the identifiability of the investigated system and on the power of the available computational facilities, uses the advantages of the appropriate method.