Background Medication adherence, one of the most important aspects in the process of optimal medicines use, is unfortunately still a major challenge in modern healthcare, and further research is required into how adherence can be assessed and optimised. The aim of this study was to use a combined method approach of self-report and dried blood spot (DBS) sampling coupled with population pharmacokinetic (PopPK) modelling, to assess adherence to metformin in adult patients with type 2 diabetes. Further aims were to assess metformin exposure levels in patients, determine factors associated with non-adherence with prescribed metformin, and to explore the relationship between adherence and therapeutic outcomes. Methods A combined method approach was used to evaluate metformin adherence in patients with type 2 diabetes who had been prescribed metformin for a minimum period of 6 months. Patients were recruited from consultant-led diabetic outpatient clinics at three hospitals in Northern Ireland, UK. Data collection involved self-reported questionnaires [Medication Adherence Report Scale (MARS), Beliefs about Medicines Questionnaire and Centre for Epidemiologic Studies Depression Scale], direct measurement of metformin concentration in DBS samples, and researcher-led patient interviews. The DBS sampling approach was coupled with population pharmacokinetic (PopPK) modelling, which took account of patient characteristics, metformin dosage and type of formulation prescribed (immediate or sustained release). Results The proportion of patients considered to be adherent to their prescribed metformin, derived from self-reported MARS scores and metformin concentration in DBS samples, was 61.2% (74 out of 121 patients). The majority (n = 103, 85.1%) of recruited patients had metformin exposure levels that fell within the therapeutic range. However, 17 patients (14.1%) had low exposure to metformin and one patient (0.8%) had undetectable metformin level in their blood sample (non-exposure). Metformin self-administration and use of a purchased adherence pill box significantly increased the probability of a patient being classified as adherent based on logistic regression analysis. Both HbA1c and random glucose levels (representing poor glycaemic control) in the present research were, however, not statistically linked to non-adherence to metformin (P > 0.05). Conclusions A significant proportion of participating patients were not fully adherent with their therapy. DBS sampling together with the use of a published PopPK model was a useful, novel, direct, objective approach to estimate levels of adherence in adult patients with type 2 diabetes (61.2%).