Attorney General Eric H. Holder Jr. said Friday that he opposes the use of data analysis in criminal sentencing, a practice that has been adopted by several states but that critics say could result in unfairly harsh sentences for minority defendants.

In a speech in Philadelphia to the National Association of Criminal Defense Lawyers, Holder said that while information on education levels, socioeconomic backgrounds and neighborhoods can be useful in some areas of law enforcement, he cautioned against using such data to determine prison sentences.

Sentencing decisions based on “static factors and immutable characteristics,” Holder said, “may exacerbate unwarranted and unjust disparities that are already far too common in our criminal justice system and in our society.”

Holder also asked the U.S. Sentencing Commission, an independent agency that establishes sentencing policies for the federal courts, to study and issue policy recommendations about the use of “big data” in sentencing decisions.

“Criminal sentences must be based on the facts, the law, the actual crimes committed, the circumstances surrounding each individual case and the defendant’s history of criminal conduct,” Holder said. “They should not be based on unchangeable factors that a person cannot control, or on the possibility of a future crime that has not taken place.”

Holder’s request to the commission is his latest action on criminal justice reform, an effort he launched last summer when he announced the department would no longer charge low-level, nonviolent drug offenders with offenses that impose severe mandatory sentences.

Last month, the commission issued a ruling — endorsed by Holder — that will allow nearly 50,000 federal drug offenders in prison to be eligible for reduced sentences. The commission made retroactive an earlier change that had lightened potential punishments for most future drug offenders who are sentenced starting in November.

Several states, including Pennsylvania and Virginia, are now using data to conduct “risk assessments” before handing down criminal sentences. Supporters say that data about a person’s past help judges make better sentencing determinations.

The Senate Judiciary Committee has passed bipartisan legislation that would require offenders to undergo risk assessments to determine whether they present a low, medium or high risk of recidivism. The bill, introduced by Sens. John Cornyn (R-Tex.) and Sheldon Whitehouse (D-R.I.), would also allow certain inmates to reduce their sentences through a combination of factors.

In a report to the commission, the Justice Department said basing criminal sentences on data is a “dangerous concept.”

The practice “will become much more concerning over time as other far reaching sociological and personal information unrelated to the crimes at issue are incorporated into risk tools,” the department said.