It started with a pit bull. Frederick Schauer, who teaches a course on the first amendment at the Kennedy School of Government at Harvard, was reading about some dog lovers who claimed "canine racism" in response to measures to curb attacks by pit bulls in New York City. That particular race card, he said, was an extreme example of how society has become so obsessed with avoiding any stereotypes that it ignores reality. Pit bulls are more aggressive than other breeds, he said, just as statistics show older people have slower reflexes than the young, and there are more bad drivers in Massachusetts than in Vermont. A fair number of generalizations, he insists, turn out to be accurate.

Amid new concerns about identifying terrorists and old worries about the police using race and ethnicity as grounds for suspicion, Mr. Schauer says one of his goals is to fine-tune "profiling" and rescue it from being pejorative. "I defend the morality of decisions by categories and by generalizations," he writes.

That is clearly no easy task. Racial profiling, for instance, is one of the thorniest public policy issues around these days. Some states, like New Jersey, have forbidden the police to use racial profiling. Last summer the White House tried to tiptoe through the politically treacherous issue by announcing guidelines to keep federal agents from using race or ethnicity in routine investigations but allowing it for terrorist or national security matters. Suspect descriptions are O.K., but race-based profiles are not, a distinction without a difference in the eyes of many scholars.

Lawyers, philosophers and others have long pondered the legal and moral distinctions among discrimination, stereotyping and statistical probabilities, but they have not reached broad agreement. Beyond suspicions about rounding up the usual suspects, how do we know whether generalizations are based on sound empirical information or are a jumble of pop culture shorthand and bad science? What kinds of broad judgments are right and what kinds are wrong? How are categories constructed in the first place?

"It's sort of human to make categories, it's a shorthand for behavior," said Cynthia Fuchs Epstein, a professor of sociology at the Graduate Center of the City University of New York. "The problem is the drawing of boundaries is arbitrary." She continued, "Stereotypes can come from social science that can come from inadequate data."

This is the kind of debate that Mr. Schauer is hoping to foster with his new book, "Profiles, Probabilities and Stereotypes" (Harvard). Discrimination is not always bad, he insists. Society discriminates all the time, when it decides at what age people can drink and vote and who should pay higher insurance premiums. Employers discriminate when they favor applicants with good grades from an elite school. Because human beings are not computers, he explains, we rely on generalizations.

Mr. Schauer stressed that he did not support bias of any kind and did not believe that empirical evidence alone should be the driving force behind creating categories. Rather, he said in an interview, he longs for a more sophisticated appreciation of the difference between generalizations based on information that helps society make rules and laws and those generalizations that are spurious or harmful. The sorting can be done, he said, by taking into account history, morality and the social impact of the generalizations.

"We've learned too much from history and too much from social science to assume we look at all categories the same," Mr. Schauer said. "One of the nice things about generalizations is that they are frequently temporal and based on real data and are out there for the public to inspect."

For instance, using race in creating a profile of a criminal, he said, is far different from the police randomly stopping and harassing black motorists on the highway. One use is legitimate, one use constitutes bias.

Cass R. Sunstein, a professor of law at the University of Chicago, said he bought Mr. Schauer's argument that life was full of stereotypes and generalizations. Still, he added, lawyers also acknowledge that rules are error-prone. Laws inevitably overshoot and undershoot. A 65-mile-an-hour speed limit, for example, covers people driving 70 on sunny days and people doing 64 on an icy road. And that's a far cry from the damage done, he said, when profiles that ostensibly use race as one factor become ways to justify racial bias.

"One danger is the self-fulfilling prophecy," Mr. Sunstein said. If discriminatory law enforcement makes it more likely that one group (in this case, African-Americans) is disproportionately imprisoned, it reinforces an inaccurate profile. "Look at the profiling done based on the percentage of a group in the prison population," he said.

The important question is, Ms. Epstein said, for what purpose do we generalize? Is it to legitimately order the way we live or to keep some people in control? "Whether a person is mature at 18 or 21 is a matter of expediency," she said.

Too many statistics are derived from bad research, Ms. Epstein added, and statistics can be variously interpreted. Her book "Deceptive Distinctions: Sex, Gender and the Social Order" (Yale University Press, 1990) argues that much gender research reflects social control and ideology more than science. "When you say men on average make more money than women, it could be that women work part time or they work fewer hours or they face discrimination," she said.

"It's very difficult because stereotypes can come from social science that can come from inadequate data. Are women kinder and more nurturing than men?" Ms. Epstein said. "There's a constant amount of social control that goes into maintaining those characteristics. There is a lot of peer pressure to conform to norms of behavior."

In the end, it is a matter of balance, say most scholars who have studied the issue. But even balance is a judgment call. In his book, Mr. Schauer tries to wrestle with a few examples of tricky balancing acts. He asks, for example, whether women should be excluded from a military academy because women, on average, are more sensitive to hazing than men. No, he argues, because the social impact of that kind of discrimination is not worth the price of keeping out some sensitive women.

Randall Kennedy, a professor of law at Harvard, notes that the law sets out guidelines for how to balance competing interests. It is already mindful, for instance, of how different categories have historically been used, that race and gender "have been used in a bad way," he said.

That is not the same as saying that race or gender don't exist, that they don't matter, or that one should never generalize, Mr. Kennedy said.

He agrees with Mr. Schauer that discrimination is not always negative, it simply denotes paying attention to some salient characteristic. Affirmative action, he said, is a type of "racial discrimination" that he supports. A lot of things go into the mix: How big are the categories? Are the characteristics immutable or temporary?

"Schauer is pressing people to think about the language that they use, the logic that they use," Mr. Kennedy said. "If you went out on the street and asked people, `What's bad about discrimination?,' they would know what they're supposed to say. But have they thought about it?"

Still, despite Mr. Schauer's intentions, profiling may be a term that has entered the lexicon in a particular way and will not go away, even if it conflates a legitimate law enforcement practice with discriminatory behavior or excessive force, said Tracey Meares, a professor of law and director of the Center for Studies in Criminal Justice at the University of Chicago.

Although the police can theoretically come up with detailed profiles of criminal suspects, she said, people nonetheless assume that the word profile means that officers use nothing but race or nationality in going after a suspect.

To Sanford Levinson, a law professor at the University of Texas, the law will never be a perfect guide to when it is acceptable to generalize. "The notion of equality is the great mystery of American public culture," Mr. Levinson said. "We veer back and forth between saying all of us ought to be treated as individuals and the realization that that doesn't make much sense."

Copyright 2003 The New York Times Company. Reprinted from The New York Times, Arts & Ideas, of Saturday, December 13, 2003.