One sample analyzed was drawn from persons convicted and sentenced to prison. The second sample contained all persons convicted, whether or not they were sentenced to prison. The third sample consisted of all persons whose cases went to the prosecutor, whether or not they were ultimately prosecuted. Legitimate and nonlegitimate influences on sentence length were investigated for each analysis. Data were obtained for all processing histories of felony arrests reaching dispositions in 1977-79. Aside from the sample selection criteria, the models tested were identical. The exogenous variables were race/ethnicity, gender, age, offense type, evidence against the defendant, offense severity, mode of disposition, number of days between arrest and disposition, and degree of crime specialization (for two or more arrests). Results indicate that bias is introduced when prosecutorial case dispositions and nonprison sentences are ignored. Also, the effects of certain exogenous variables on sentence length differ, depending on the sample selection criteria. Of these, the influence of pleading guilty rather than going to trial is particularly significant. Overall, the findings are consistent with the conclusion that earlier sentencing studies have concealed the effects of sentence bargaining because of bias in sample selection. Tabular data and 74 reference listings are provided.