Circumstantial evidence, observed Sherlock Holmes, “is a very tricky thing. It may seem to point straight to one thing, but if you shift your own point of view a little, you may find it pointing in an equally uncompromising manner to something entirely different.”

I’m willing to bet that Holmes would have been just as cautious about the use and misuse of DNA testing. The more we learn about a technique that was once hailed as the panacea of criminal investigation, the more we realise how far it is from infallibility. As Holmes might have said: “There is nothing more deceptive than a single piece of evidence upon which an entire case is said to rest.”

This doubt alone should be reason enough to steer clear of moves to store the DNA of all crime suspects, however minor, or even to include the entire population, on a national database. Ethics, civil liberties, the threat of a police state – all of these are powerful arguments against it. Ultimately, however, what tilts the balance are the questions that surround the reliability of DNA itself. We know very little of the forensic evidence in the McCann case, but what seems clear enough is that the DNA samples gathered from the apartment and the hire car used by Gerry and Kate McCann are not, of themselves, sufficiently convincing to launch a prosecution. Given the obvious suspicions of the Portuguese police, one can only deduce that, had the DNA links stood up, the McCanns would already be on their way back to face further questioning.

The value of that evidence may be no better than the DNA analysis that failed so spectacularly last week in the World’s End murder trial. It was the heart and soul of the state prosecution case, advanced after months of expert analysis, yet thrown out after only two weeks by a High Court judge who ruled that the case was not strong enough to go before a jury.

The brutal murder, 30 years ago in Edinburgh, of two 17-year-old girls – Christine Eadie and Helen Scott – was resurrected for two reasons that, when brought together, make a powerful case for action, but do not necessarily further the interests of justice. The families of both girls understandably wanted the killer or killers to be found, tried and convicted. The police believed that advances in forensic science might give them the evidence they lacked 30 years ago.

They had a suspect, Angus Sinclair, already serving time for another murder, and they had at their disposal the results of some brilliant laboratory tests, which brought Sinclair tantalisingly close to the scene of the murder.

It was perhaps this brilliance that was their undoing. The forensic team borrowed the highly sophisticated technique of “ancestral” DNA testing – used recently by London detectives to identify the origins of a small African boy’s torso found in the Thames – to trace Sinclair’s genetic “fingerprint” in a semen stain left on Helen Scott’s coat. They discovered elements of DNA in the foam mouldings used by Sinclair’s brother-in-law to redecorate a flat. Their investigation brought forward an irresistible case to demonstrate that there had been sexual contact between both men and the two girls.

The DNA evidence fell apart, however, when it came to proving that, as well as having sex with the girls, Sinclair had murdered them.

Both had been strangled, and the ligatures used to kill them were tested, using an ultra-sensitive low-copy DNA profiling technique. The results of this test were described in a forensic report as “moderately strong evidence” of what might have been expected if Sinclair had been involved in tying the knots. But before it could even be put to the judge, the prosecuting counsel withdrew it. By that stage it had become clear from the judge’s comments that the case would not stand up, that evidence of abduction, rape and murder was not strong enough.

The decision has inevitably caused controversy, but when, last week, the chief prosecutor – Scotland’s Lord Advocate – came before the Scottish Parliament to explain in the grimmest of detail why the case had nevertheless been launched, she admitted that the DNA evidence amounted only to “low probability”.

Neither “moderately strong” or “low probability” sounds like cast-iron testimony. And if that difference of opinion can apply to a high-profile murder case, where experts have pored over the evidence for many months, how much more likely it is that overworked and overstretched police will get it wrong when it comes to more common cases of rape, robbery and criminal assault.

Because of this fatal imprecision, it would be wrong to continue including the DNA profile of every suspect, whether eventually found innocent or guilty, on a national database, as happens now in England, Wales and Northern Ireland. This week, the Nuffield Council on Bioethics said that it was unacceptable for police to store records taken from suspects who are later acquitted or never charged. That way, it concluded, innocent people could come under suspicion, without any great benefit to criminal justice. Downing Street, by contrast, believes that not only should the present system be maintained, it should be extended to include people suspected of minor crimes such as traffic offences and dropping litter. The gap between that and a DNA database that carries the genetic records of every citizen in the land is becoming steadily narrower.

DNA has been of huge value in solving a wide range of violent crimes, providing unexpected solutions to past mysteries and bringing comfort to the families of victims who yearn for justice. It is not, however, as reliable as we would like to believe, and it remains just one weapon in the armoury of detection rather than the magic solution to every unsolved case.

Times Recommends

• We must lead the world to stability
• The recession's a blessing in disguise
• The guilty men who sent mum to Iceland