The reliability and validity of proxy respondent information in the Canadian Hospitals Injury Reporting and Prevention Program surveillance system was assessed. A standardized form was used to collect data on injury date, location, context (activity at the time), breakdown factor (what went wrong), mechanism, product involvement, safety precaution use, and motor vehicle involvement. The test-retest method determined reliability, with the kappa coefficient quantifying agreement between respondent information provided in the emergency department and later during a telephone interview. Of 421 eligible respondents, 325 (77%) completed the telephone interview, with a median time to interview of 33 hours (range 24-70 hours). Agreement was high for all items; kappa coefficients ranged from 0.79 (substantial agreement) to 1.00 (perfect agreement). Reliability was not significantly modified by respondent view of the injury event, age of the child, language of the form, or level of respondent education. Validity was determined by measuring the agreement between respondent information and that provided by an independent witness. Witness information was considered to represent the truth. Of the 140 injury events selected, 92 (66%) had the form completed by both the original respondent and an independent witness. Kappa coefficients were greater than 0.65 for all but one item (safety precaution use), and the positive predictive value of respondent information for item categories whose prevalence was > or = 0.25 ranged from 0.82 to 0.95. The authors conclude that proxy respondent data on childhood injury are both reliable and valid.