The Uniform Crime Reporting (UCR) system was founded in 1929 with an organizational design intended to encourage compliance with its rules for counting reported crimes and arrests; however, the complexity and ambiguity of the counting rules have contributed to statistical unreliability in arrest records. Some of the difficulties lie in differences in definitions and crime classifications across States, differences in arrest definitions, and the problem of classifying incidents that involve multiple crimes. An empirical analysis of compliance with counting regulations involved site visits of 18 police agencies to interview recordkeeping staff and observe arrest operations. Based on the issues that emerged in the site visits, a survey was developed and mailed to all 213 city, county, and sheriffs' departments identified as serving populations of 100,000 or more, plus a random sample of 26 agencies serving populations of 10,000 - 100,000. A total of 196 supplied usable responses. The site visits also produced a set of questions for the State agencies receiving the local police statistics. Four police agencies were selected for 2-week site visits. The findings show that the State UCR agencies invest relatively little effort in regulating arrest statistics and that the regulators themselves often fail to comply with UCR counting rules. This weak regulatory system allows a fairly high rate of errors on certain issues in the local police agencies. Most important was the error rate within agencies for certain offenses, which was found to be quite high. Substantive implications of these findings for arrest analyses, policy implications, and general implications for social indicators are discussed. Tabular data are provided.