The majority of environmental models require calibration of some or all of their parameters before meaningful predictions of catchment behaviour can be made. Despite the importance of reliable parameter estimates, there are growing concerns about the ability of objective-based inference methods to adequately calibrate environmental models. The problem lies with the formulation of the objective or likelihood function, which is currently implemented using essentially ad-hoc methods. We outline limitations of current calibration methodologies, including least squares, multi-objective, GLUE and Kalman filter schemes and introduce a more systematic Bayesian Total Error Analysis (BATEA) framework for environmental model calibration and validation. BATEA imposes a hitherto missing rigour in environmental modelling by requiring the specification of physically realistic uncertainty models with explicit assumptions that can and must be tested against available evidence. Distinguishing between the various sources of errors will reduce the current ambiguity about parameter and predictive uncertainty and enable rational testing of environmental model hypotheses. A synthetic study demonstrates that explicitly accounting for forcing errors leads to immediate advantages over traditional least squares methods that ignore rainfall history corruption and do not directly address the sources of uncertainty in the calibration. We expect that confronting all sources of uncertainty, including data and model errors, will force fundamental shifts in the model calibration/verification philosophy.