This chapter considers the evaluation of forecasts. It begins with the evaluation of forecast performance based on the efficient use of information, for both rolling-event and fixed-event forecasts. Notions of efficiency and unbiasedness are viewed as desirable properties for forecasts to possess, but, in practice, comparisons between forecasts produced by rival models would seem to be a preferable alternative to evaluation based on a comparison of individual-model forecasts with outcomes. This requires a metric for assessing forecast accuracy. Although contextspecific cost functions defining mappings between forecast errors and the costs of making those errors may exist in some instances, general MSFE-based measures have been the dominant criteria for assessing forecast accuracy in macroeconomic forecasting. However, such measures typically are not invariant to non-singular, scale-preserving linear transforms, even though linear models are. Consequently, different rankings across models or methods can be obtained from various MSFE measures by choosing alternative yet isomorphic representations of a given model. Thus, MSFE rankings can be an artefact of the linear transformation selected. A generalized forecast-error second-moment criterion is proposed with the property of invariance, but interesting problems relating to model choice and the forecast horizon remain.