Objective: To evaluate the efficacy of an exercise intervention, in addition to usual care, to prevent or delay insulin prescription in women with gestational diabetes mellitus (GDM). Design: Randomised controlled trial Setting: University hospital Population: Women at 25-35 weeks pregnancy upon diagnosis of GDM. Methods: In the intervention arm, women had weekly, supervised, 30-45 min long exercise sessions and were encouraged to accumulate more than 5000 steps per day, monitored by a pedometer, in addition to usual care. Women in the control group had usual care only. Main outcome measure: Insulin prescription. Results: From February 2008 through April 2013, 109 women were randomized into intervention (n=57) or usual care (n=52). Two women, both in the intervention group, were excluded from the analysis (one randomised in error and one lost to follow-up). Six women never attended the exercise sessions and two women participated to less than two sessions, while two-third of women were considered as compliant to the intervention. Incidence of insulin prescription was not significantly different between groups: 31 (56%) in the intervention and 24 (46%) in the control group (RR 1.22, 95% CI 0.84 to 1.78); the median time between randomization and insulin prescription was not different between groups (14 and 13 days, respectively). Conclusion: We were unable to show that an exercise program reduces or delays insulin prescription. Low compliance with the intervention, small sample size and the short duration of the program may explain the apparent lack of benefit.