Evaluating appropriate error measures to determine demand forecast accuracy is essential in model selection, however there is no approach that simultaneously evaluates different model classes and several inter-dependent error measures. Furthermore, error measures may yield conflicting results making it more difficult to select the ‘best’ forecasting model when considering several error measures simultaneously. This paper proposes a novel process of evaluation of demand forecasting models using the analytical network process combined with the technique for order of preference by similarity to ideal solution (ANP-TOPSIS) which incorporates interdependence amongst error measures. The methodology is validated through an implementation case of a plastic bag manufacturer demonstrating that the use of the ANP-TOPSIS approach, avoided the selection of an inappropriate forecasting model due to conflicting error measurements. Moreover, a sensitivity analysis finds that the interdependence between the error measures is found to impact the relative closeness to the ideal solution, even though it plays a minimal role in the final ranking of the forecasting models.