particular set of candidates for whom arrest and crime rates might be expected to be predictive is publicly elected law enforcement officials. Using 1976 Florida County Sheriff election data, this article examines the capacity for discriminating between winning and losing incumbent sheriffs by using county arrest and crime data in association with socioeconomic and political information. The findings show that a significant discrimination is obtained (significant = 0.009, canonical correlation = 0.59, and 76 percent of the counties were correctly classified) and that crimes and arrests are significant factors in determining the outcome of sheriff elections. The results indicate that although traditional political factors, such as party affiliation and number of terms in office, and socioeconomic factors, such as income, density, and unemployment rates, do well in discriminating winning from losing candidates, a sheriff-election model must also incorporate crime and arrest information, particularly information on murder and rape. (Author abstract)