For a sample of 25 cities, this paper computes and compares race-specific offending rates for robbery and assault from two alternative data sources: victims' reports of offenders' race, as recorded in the National Crime Survey's Cities Sample, and official arrest data, as compiled by the FBI. In support of Sellin's argument that each step in the disposition of a crime introduces selective biases, the convergence between the total crime rates computed from these two sources is slightly weaker than that previously reported for the NCS crime rates and crime rates based on the UCR "offenses known to the police." The observed correlations are stronger for blacks than for whites, especially for the offense of robbery. Finally, two structural characteristics of cities, income inequality and population density, are found to suppress the bivariate association between the victim-reported (NCS) crime rates and the arrest rates. 4 tables and 34 references. (Author abstract)