Aims: Delivering adequate diabetes care is difficult in rural Africa because of drug and equipment shortages; as well as lack of trained medical expertise. We aimed to set up and evaluate a nurse-led protocol and education-based system in rural Kwazulu Natal in South Africa. Methods: A treatment algorithm and education system adapted from previously validated methods was used; care was devolved to primary health clinics and was delivered by two nurses. Glycaemic control was assessed by glycated haemoglobin (HbA 1c), assayed off site and not available for clinical use during the study. Results: A total of 284 patients were enrolled, with 197 followed for 18 months (13 died and 26% lapsed during the period). HbA1c was 11.6 ± 4.5% (sd) at baseline, 8.7 ± 2.3% at 6 months and 7.7 ± 2.0% at 18 months. There was a small associated increase in weight but no increase in hypoglycaemia. Subgroup analysis showed that education alone, without drug type or dose changes, also improved control (HbA1c 10.6 ± 4.2% baseline and 7.6 ± 2.3% at 18 months). The service was very popular with patients, families and other health workers. Conclusions: We conclude that a simple protocol and education-based diabetes care system can be successfully introduced and run by nurses in rural Africa. Medium-term glycaemic improvements are excellent and the service has been very well received.