Ingmar Guandique, center, accused of killing Washington intern Chandra Levy, is escorted from the Violent Crimes Unit in Washington by detectives Tom Williams, left, and Emilio Martinez in April 2009. Prosecutors recently dropped charges against him. (Jacquelyn Martin/Associated Press)

THERE WAS no DNA, no forensics, no eyewitnesses linking Ingmar Guandique to the death of Chandra Levy. Nonetheless, a jury in 2010 convicted him of murder, largely because of the testimony of a jailhouse informant. So when questions arose about the credibility of that witness, the case crumbled, with a judge first ordering a new trial and prosecutors eventually deciding to dismiss the charges. The events underscore the vulnerability — critics say unreliability — of such witnesses, and that should prompt law enforcement officials to reassess their use.

In what amounts to the latest wrinkle in a mystery that has confounded the country since 2001, when the 24-year-old intern went missing in the District, prosecutors last week asked a judge to dismiss the case against Mr. Guandique. Convicted in 2010, Mr. Guandique was set for retrial this fall when the surprise announcement was made. Citing “recent unforeseen developments,” the U.S. Attorney’s Office said it concluded “it can no longer prove the murder case against Mr. Guandique beyond a reasonable doubt.” The move came after a woman gave authorities recordings she had secretly made with the government’s star witness, a former cellmate of Mr. Guandique who claimed Mr. Guandique had confessed to Levy’s murder.

It is unclear what the witness said on the tapes that prompted authorities to pull back. Also unclear is whether prosecutors still believe that Mr. Guandique, an illegal immigrant now subject to deportation proceedings, is responsible for Levy’s death; if they do, their dismissal of the case raises a whole new raft of troubling questions. What is apparent is the risk involved in relying so heavily on someone who, as defense attorneys contended, may have had a motive to fabricate testimony.

No data exists on how often jailhouse informants factor into cases, but University of Virginia School of Law professor Brandon L. Garrett said they are used with “some frequency.” His research examining the first 330 DNA exonerations in the United States found that 22 percent of the wrongful-conviction cases examined had informant testimony, including jailhouse informants and other types of incentivized witnesses. Because of the problems inherent with jailhouse informants, some think using them should be banned, while others advocate reforms, being pioneered in a few states, for stringent judicial review or videotaping of interrogations.

And it’s not just the accused who can be hurt when something goes wrong with a jailhouse informant. Just ask Levy’s parents, who are still waiting — after 15 years — for justice in their daughter’s death.