The value of modelling for the study of accidents and for devising and assessing methods of prevention depends on the reliability of data sources. Police statistics are commonly used for accidents which cause death or injury. Several studies, however, show their deficiencies particularly for injury accidents of pedestrians and pedal cyclists. Hospital records give better information on injuries and diagnostic classification by the ICD code can be usefully improved by inclusion of the "E" (circumstance) detail. Insurance data can provide another source which is likely to be best for accidents involving multiple vehicles and damage to the property of third parties. The special "no fault" compensation system in New Zealand gives an unusually full source of data suitable for statistical analysis of accidents.