The effort to develop operational definitions of high-rate serious criminals used data from the RAND Corporation's Second Inmate Survey. The survey gathered data from 956 male inmates in county jails and State prisons in California, Michigan, and Texas in 1978 and 1979. Logistic regression equations were used to estimate the probability that an offender was high-rate serious, based on any of six possible definitions, as a function of the official record characteristics. None of the definitions distinguished clearly between the most serious offenders and the others. However, the best discriminant rule using all self-report data had a success rate about 40 percent higher than a rule based on chance. Thus, using discriminant rules with carefully recorded juvenile record information may improve the identification of high-rate serious offenders. Data tables, footnotes, appended tables and survey instruments, and 58 references.