Sentencing guidelines are said to be empirically derived from data on past paroling or sentencing decisions, but most models have considered only selected factors and their results have been modified to reflect subjective policy decisions. One reason that past practice is not fully described in a sentencing guidelines model is that it may have contained variations caused by items considered inappropriate or unethical. Thus, race and sex may be excluded even though they may be highly correlated with other items treated as acceptable. Several methods have tried to remove the effects of inappropriate factors from sentencing guidelines models, including simply ignoring them and examining the relations that discriminatory variables have with other items in the data. This study examines the last approach, using data on sentences imposed on 1,440 defendants in Massachusetts during 1977-78 and factors included in the State's sentencing guidelines model -- seriousness of current offense, seriousness of any prior offense, amount of injury to victim, and whether the offender used a weapon in the current offense. Massachusetts judiciary and sentencing guidelines staff were concerned about racial disparity, but their research project reported that a defendant's race did not directly influence sentence length, although the variables that influenced sentence length were weighted differently for black and white defendants. This study determined that race did affect sentence lengths in Massachusetts, accounting for an additional 5 months in the sentences of black offenders. The data also suggested that seriousness of prior record was an important consideration for judges when sentencing white offenders, but did not matter much in the sentence length for black offenders. Exclusion of race and other such factors from modeling equations may thus build into the subsequent guidelines the very kinds of things they are supposed to avoid. Tables, 16 footnotes, and 21 references are included. See NCJ-87442.