Many children with diabetes require small doses of insulin administered with syringes or pen-injector devices (at the Booth Hall Paediatric Diabetic Clinic, 20% of children aged 0-5 years receive 1-2 U insulin doses). To determine how accurately and reproducibly small doses are delivered, 1, 2, 5, and 10 U doses of soluble insulin (100 U/ml) were dispensed in random order 15 times from five new NovoPens (1.5 ml), five BD-Pens (1.5 ml), and by five nurses using 30 U syringes. Each dose was weighed, and intended and actual doses compared. The two pen-injectors delivered less insulin than syringes, differences being inversely proportional to dose. For 1 U (mean (SD)): 0.89 (0.04) U (NovoPen), 0.92 (0.03) U (BD-Pen), 1.23 (0.09) U (syringe); and for 10 U: 9.8 (0.1) U (NovoPen), 9.9 (0.1) U (BD-Pen), 10.1 (0.1) U (syringe). The accuracy (percentage errors) of the pen-injectors was similar and more accurate than syringes delivering 1, 2, and 5 U of insulin. Errors for 1 U: 11(4)% (NovoPen), 8(3)% (BD-Pen), 23(9)% (syringe). The reproducibility (coefficient of variation) of actual doses was similar (< 7%) for all three devices, which were equally consistent at underdosing (pen-injectors) or overdosing (syringes) insulin. All three devices, especially syringes, are unacceptably inaccurate when delivering 1 U doses of insulin. Patients on low doses need to be educated that their dose may alter when they transfer from one device to another.