Diabetes mellitus is a major health risk in many countries, and the incidence rates are increasing. Diverse therapeutic agents are applied to treat this condition. Since 1960, numerous mathematical models have been developed to describe the glucose-insulin system, analyse data from diagnostic tests and quantify drug effects. This review summarizes the present state-of-the-art in diabetes modelling, with a focus on models describing drug effects, and identifies major strengths and limitations of the published models. For diagnostic purposes, the minimal model has remained the most popular choice for several decades, and numerous extensions have been developed. Use of the minimal model is limited for applications other than diagnostic tests. More mechanistic models that include glucose-insulin feedback in both directions have been applied. The use of biophase distribution models for the description of drug effects is not always appropriate. More recently, the effects of various antidiabetic agents on glucose and insulin have been modelled with indirect response models. Such models provide good curve fits and mechanistic descriptions of the effects of antidiabetic drugs on glucose-insulin homeostasis. These and other types of models were used to describe secondary drug effects on glucose and insulin, and effects on ancillary biomarkers. Modelling of disease progression in diabetes can utilize indirect response models as a disturbance of homeostasis. Future needs are to include glucose-insulin feedback more often, develop mechanistic models for new drug groups, consider dual drug effects on complementary subsystems, and incorporate elements of disease progression.