Treatment-related mortality is an important outcome in paediatric cancer clinical trials. An international group of experts in supportive care in paediatric cancer developed a consensus-based definition of treatment-related mortality and a cause-of-death attribution system. The reliability and validity of the system was tested in 30 deaths, which were independently assessed by two clinical research associates and two paediatric oncologists. We defined treatment-related mortality as death occurring in the absence of progressive cancer. Of the 30 reviewed deaths, the reliability of classification for treatment-related mortality was noted as excellent by clinical research associates (κ=0·83, 95% CI 0·60-1·00) and paediatric oncologists (0·84, 0·63-1·00). Criterion validity was established because agreement between the consensus classifications by clinical research associates and paediatric oncologists was almost perfect (0·92, 0·78-1·00). Our approach should allow comparison of treatment-related mortality across trials and across time.