Aims/hypothesisThe study aimed to compare the effects of a 2 year intervention with a low-fat diet (LFD) or a low-carbohydrate diet (LCD), based on four group meetings to achieve compliance.MethodsThis was a prospective randomised parallel trial involving 61 adults with type 2 diabetes consecutively recruited in primary care and randomised by drawing ballots. Patients that did not speak Swedish could not be recruited. The primary outcomes in this non-blinded study were weight and HbA1c. Patients on the LFD aimed for 55–60 energy per cent (E%) and those on LCD for 20 E% from carbohydrate.ResultsThe mean BMI and HbA1c of the participants were 32.7 ± 5.4 kg/m2 and 57.0 ± 9.2 mmol/mol, respectively. No patients were lost to follow-up. Weight loss did not differ between groups and was maximal at 6 months: LFD −3.99 ± 4.1 kg (n = 31); LCD −4.31 ± 3.6 kg (n = 30); p < 0.001 within groups. At 24 months, patients on the LFD had lost −2.97 ± 4.9 kg and those on LCD −2.34 ± 5.1 kg compared with baseline (p = 0.002 and p = 0.020 within groups, respectively). HbA1c fell in the LCD group only (LCD at 6 months −4.8 ± 8.3 mmol/mol, p = 0.004, at 12 months −2.2 ± 7.7 mmol/mol, p = 0.12; LFD at 6 months −0.9 ± 8.8 mmol/mol, p = 0.56). At 6 months, HDL-cholesterol had increased with the LCD (from 1.13 ± 0.33 mmol/l to 1.25 ± 0.47 mmol/l, p = 0.018) while LDL-cholesterol did not differ between groups. Insulin doses were reduced in the LCD group (0 months, LCD 42 ± 65 E, LFD 39 ± 51 E; 6 months, LCD 30 ± 47 E, LFD 38 ± 48 E; p = 0.046 for between-group change).Conclusions/interpretationWeight changes did not differ between the diet groups, while insulin doses were reduced significantly more with the LCD at 6 months, when compliance was good. Thus, aiming for 20% of energy intake from carbohydrates is safe with respect to cardiovascular risk compared with the traditional LFD and this approach could constitute a treatment alternative.Trial registration:ClinicalTrials.gov NCT01005498Funding:University Hospital of Linköping Research Funds, Linköping University, the County Council of Östergötland, and the Diabetes Research Centre of Linköping UniversityElectronic supplementary materialThe online version of this article (doi:10.1007/s00125-012-2567-4) contains peer-reviewed but unedited supplementary material, which is available to authorised users.