Findings show that the SHR data have several error sources, including incomplete information, frequent police assumptions that a homicide is in the most common category, and apparent changes over time in decision rules regarding classification. However, even with the errors and inconsistencies in homicide classification, it appears clear that more stranger murders and more murders in connection with some other offense are occurring than are indicated by the SHR data. The declining numbers of conflict homicides and other felony-related homicides and the increase in the numbers classified as unknown support this conclusion. Findings show the usefulness of the data for explaining variations in the risk of becoming a homicide victim across age groups. Results also suggest that future research needs to focus closely on how categories are used and on reliability and validity problems in SHR data. Figures, tables, footnotes, and 40 references.