Self-care is essential for patients with diabetes mellitus. Both clinicians and researchers must be able to assess the quality of that self-care. Available tools have various limitations and none are theoretically based. The aims of this study were to develop and to test the psychometric properties of a new instrument based on the middle range-theory of self-care of chronic illness: the Self-Care of Diabetes Inventory (SCODI). Forty SCODI items (5 point Likert type scale) were developed based on clinical recommendations and grouped into 4 dimensions: self-care maintenance, self-care monitoring, self-care management and self-care confidence based on the theory. Content validity was assessed by a multidisciplinary panel of experts. A multi-centre cross-sectional study was conducted in a consecutive sample of 200 type 1 and type 2 diabetes patients. Dimensionality was evaluated by exploratory factor analyses. Multidimensional model based reliability was estimated for each scale. Multiple regression models estimating associations between SCODI scores and glycated haemoglobin (HbA1c), body mass index, and diabetes complications, were used for construct validity. Content validity ratio was 100%. A multidimensional structure emerged for the 4 scales. Multidimensional model-based reliabilities were between 0.81 (maintenance) and 0.89 (confidence). Significant associations were found between self-care maintenance and HbA1c (p = 0.02) and between self-care monitoring and diabetes complications (p = 0.04). Self-care management was associated with BMI (p = 0.004) and diabetes complications (p = 0.03). Self-care confidence was a significant predictor of self-care maintenance, monitoring and management (all p < 0.0001). The SCODI is a valid and reliable theoretically-grounded tool to measure self-care in type 1 and type 2 DM patients.