The research tested the liberation hypothesis, which suggests that the greater the offense severity, the less likely judges or juries are to feel free to follow their own sentiments regarding guilt and punishment. As a result, this hypothesis also suggests that the ability of legal variables to predict variation in sentence length will be greater as the severity of the crime increases. This theory contrasts with structural-contextual theory, which suggests that the components of the justice system traditionally work somewhat independently of one another and the explained variance in sentences will increase appreciably when components function with greater interdependence. The samples included 111 terrorists listed in Federal Bureau of Investigation records and indicted in 1982-89, as well as 482 matched nonterrorists from the Federal court database of the Federal Judicial Center. Data were analyzed by means of least-squares regression and structural equation modeling procedures. Results revealed strong support for both theories. Explained variance for the terrorists was more than four times greater than the explained variance for the nonterrorists. Further analysis revealed that explained variance was highest for terrorists who committed offenses of high severity and lowest for nonterrorists who committed offenses of low severity. The subsequent addition of other predictor variables available only for the terrorist sample further increased the explained variance and provided additional support for the liberation hypothesis. Figures, tables, footnotes, appended methodological information, and 38 references