For goal-oriented model adaptation a model-error estimator is required to drive the adaptation process. In recent years publications have appeared on the dual-weighted residual (DWR) method in the application of model-error estimation in output functionals. In this paper we study the application of the DWR method for convection-diffusion problems where hierarchical models are of different type. Omitting the diffusion operator often results in a singular perturbation problem considering the model residual in the limit of vanishing diffusion. This is caused by the change of mathematical type of the model equations and therefore the applied boundary conditions. In this work we show how a model error estimator is developed for steady and unsteady convection-diffusion problems. It is found that a weak formulation and weakly imposing boundary and initial conditions leads to a dual-weighted model-error estimator that also incorporates boundary residuals.