This report documents the successive results of multi-stage research to develop, test, and field experimental Hispanic origin and race questions for a census of residential juvenile facilities. This research was done by Census Bureau staff as part of the wider Children in Custody Questionnaire Redesign Project between 1994 and 1997, with approval from both the project sponsor–the Office of Juvenile Justice and Delinquency Prevention in the Justice Department–and the Office of Management and Budget (OMB). This report was submitted to OMB during the review period of the Directive 15 race and Hispanic origin categories mandated for use in federal data collections from 1977 to 1997. This is one of a very few studies on the reporting of race and Hispanic origin in establishment data collections, where one third-person respondent typically reports on a large number of persons who may or may not be known personally to him/her. Most research has focused on improving the collection of race and Hispanic origin in the context of household surveys where one household respondent typically reports on a small number of coresident relatives and/or friends well-known to him/her. The paper starts with a description of the project background and an overview of the four stages of exploratory, cognitive, and mailout survey research to develop, test, and finalize the Hispanic origin and race questions in the new Census of Juveniles in Residential Placement facility questionnaire. A brief summary of the literature on third-person reporting of origin and race in non- household data collections is presented and the apparent lack of any previous study on the performance of these types of questions in a residential facility data collection is noted. Six general factors that may affect the quality of third-person reporting of Hispanic origin and race are identified. The mix of factors affecting the quality of race and Hispanic origin data collected from residential juvenile facilities is shown to differ from the combination of factors influencing these data in household surveys. The iterative development and testing with third-person reporters of first separate, then combined Hispanic origin and race questions in two rounds of cognitive interviews is described and the performance of a combined Hispanic origin/race question with an "other, specify" line in a mail-out test survey is then analyzed. The very high item response rate of 99.2% on more than 8,000 rostered juveniles in the unedited database before any editing and/or callbacks were started strongly suggests that the combined origin/race question was understandable to respondents and consistent with the way they collect and tabulate race data for their own files. The conclusion is that the best method for obtaining Hispanic origin and race data from third-person reporters in this establishment census is to use a combined Hispanic origin/race question, not the separate Hispanic origin and race questions recommended for use in household surveys by the Interagency Committee for the Review of the Racial and Ethnic Standards and recently adopted by OMB for data collections based on self-reports.