Transfer function models have been used for modeling the linear dependence between stationary time series and for improving the accuracy of forecasts. This article proposes a method to evaluate the fit of such a model by comparing certain frequency domain functionals of the model with those of the data using a model-based bootstrap algorithm. The method works by checking the ability of the fitted model to reproduce the dynamic relationship and the linear dependence structure of the data. Furthermore, the method addresses the problem of the possible sources of model inadequacy. The asymptotic validity of the bootstrap procedure used to evaluate the distribution of the statistics considered is proved some examples illustrating the ability of the proposed method to check the overall fit and to detect sources of model inadequacy are given.