Since they lack insulin-producing ?-cells, patients with type 1 diabetes mellitus (T1DM) need to supply their body with insulin from external sources to manage their blood glucose (BG) concentration and miti- gate the long-term effects of chronically increased BG levels. The common way of dosing insulin in T1DM is basal bolus therapy. In this method, patients continuously supply their body with a small amount of insulin that is meant for keeping their BG level more or less constant in case no large disturbances occur. Bolus insulin (which makes up roughly 30?60% of the total insulin amount) on the other hand is used to counterbalance such disturbances. The biggest perturbation of BG is caused by meals that can lead to large postprandial glucose excursions. By far, the most common approach to determine bolus insulin requirements in T1DM is known as Advanced Carbohydrate Counting (ACC). In ACC the bolus insulin amount is determined proportional to the estimated carbohydrate content of the ingested meal. Even though this semi-heuristic approach has proven very valuable in daily practice, its use is not without pitfalls. In this paper we discuss the background, implicit assumptions and limitations of ACC from an engineering perspective and show how concepts from the fields of data-based modeling and control have been successfully used to facilitate the computation of bolus insulin requirements.