Crime rates peaked nationally in 1995 and have declined substantially over the last decade. In 1995, there were 684 violent crimes nationally per 100,000 residents; in 2006, there were 473. The much-publicized decline in New York City's violent crime rate saw it go from 2,384 per 100,000 residents in 1995 to 638 in 2006. Los Angeles's violent crime rate dropped from 2,405 to 787. Smaller cities show the same trend. Kansas City's violent crime rate fell from 1,930 to 857. Even Detroit (despite its economic woes and population losses) had a slightly lower rate of violent crime in 2006 than 1995, 2,419 down from 2,699.

One place has topped all of these impressive figures, yet it has received little publicity and no credit. This crime decline was quite unexpected even, indeed especially, by criminologists. The location we have in mind is U.S. prisons. Between 1973 and 2003, the homicide rate in state prisons declined a staggering 94 percent.

Modern record keeping in prisons begins in 1925, and for the 50 years thereafter the number of state and federal prisoners kept pace with the growth of the general population. The rate of imprisonment remained stable, at about 110 inmates per 100,000 people. Society would, like a thermostat, adjust its imprisonment decisions whenever the rate of imprisonment began to fall or rise too far. Then, in the mid-1970s, the number of prisoners began shooting upward and has continued in that direction ever since. In 2006 (the last year for which statistics are available), the number of people in state and federal prisons approached 1.6 million--a rate of 501 inmates per 100,000 people. If the inmates of American prisons were assembled in a single locality, it would be the fourth largest U.S. city--tied with Philadelphia.

Whatever the value the prison buildup would have in reducing crime on the streets, reputable criminologists feared that prisons on a mass scale would have internal troubles so severe that they would lead to an organizational collapse. They predicted that mass prisons would have all the features of failed states: Such prisons would be tense, dangerous, and too weakly governed to prevent high rates of individual and collective violence. The prison riots of the 1970s and 1980s (like the 1971 debacle at Attica, which left 43 dead, or the 1980 bloodbath in Santa Fe with 33 deaths) would multiply. The only question for many criminologists was when the buildup would pass the tipping point that would precipitate mass disorder: When would the line tracking violence shoot straight up?

But the tipping point never came. In fact, the opposite occurred, with prison violence trending downward for decades. Prison crime data are notoriously problematic, but statistics on some types of crime are more reliable than others. Homicides are the best-measured crimes; they are almost always reported and accurately counted in official statistics. In 1973, there were 63 homicides per 100,000 state inmates. In 1990, there were 8, and in 2003, the homicide rate dropped further to 4. The chances of being murdered behind bars plunged during the buildup of the prison population. Prison suicide rates also dropped sharply. In 1980, there were 34 inmate suicides per 100,000 inmates. This rate decreased to 16 suicides per 100,000 in 1990, and has remained stable.

And the predicted wave of prison riots never appeared. Both the absolute number of riots and the ratio of riots to inmates declined. By 2000, prison riots had become rare. They do happen from time to time--most recently, in a private prison in Indiana housing inmates transferred there to relieve overcrowding in Arizona (no deaths, a handful of injuries)--just not very often.