Aims: Carbohydrate (CHO) counting allows children with Type 1 diabetes to adjust mealtime insulin dose to carbohydrate intake. Little is known about the ability of children to count CHO and whether a particular method for assessing CHO quantity is better than others. We investigated how accurately children and their caregivers estimate carbohydrate, and whether counting in gram increments improves accuracy compared with CHO portions or exchanges. Methods: One hundred and two children and adolescents (age range 8.3–18.1 years) on intensive insulin therapy and 110 caregivers independently estimated the CHO content of 17 standardized meals (containing 8–90 g CHO), using whichever method of carbohydrate quantification they had been taught (gram increments, 10-g portions or 15-g exchanges). Results: Seventy-three per cent (n = 2530) of all estimates were within 10–15 g of actual CHO content. There was no relationship between the mean percentage error and method of carbohydrate counting or glycated haemoglobin (HbA1c) (P > 0.05). Mean gram error and meal size were negatively correlated (r = -0.70, P < 0.0001). The longer children had been CHO counting the greater the mean percentage error (r = 0.173, P = 0.014). Core foods in non-standard quantities were most frequently inaccurately estimated, while individually labelled foods were most often accurately estimated. Conclusions: Children with Type 1 diabetes and their caregivers can estimate the carbohydrate content of meals with reasonable accuracy. Teaching CHO counting in gram increments did not improve accuracy compared with CHO portions or exchanges. Large meals tended to be underestimated and snacks overestimated. Repeated age-appropriate education appears necessary to maintain accuracy in carbohydrate estimations.