Myelodysplastic syndrome (MDS) is considered to be very rare in children. However, the only two published population‐based studies reported widely divergent incidence figures. To further explore the epidemiology of childhood MDS and to evaluate the accuracy of cancer registry and treatment trial data, we conducted a population‐based study of children aged 0–14 years in British Columbia (BC), Canada, between 1982 and 1996. MDS was diagnosed in 31 cases corresponding to an annual incidence of 3.2 per million children or 6% of all leukaemias, compared with an incidence of 6.0/million for acute myeloid leukaemia (AML), and of 0.5/million for chronic myeloid leukaemia. There was a non‐significant (P = 0.19) trend toward an increase in MDS incidence with time, the increase was partly explained by an increasing number of patients with Down syndrome. Associated abnormalities were found in 48% of the MDS cases with Down syndrome as the most common (seven cases). Only one third of the MDS cases were correctly registered in the Cancer Registry and less than half of the eligible MDS patients were enrolled on a cooperative group study. Data on MDS from treatment‐based studies and cancer registries were inaccurate and seemed to significantly underestimate the incidence of MDS in children.