"If it helps me, I don't need science to show that it works." This is the argument often voiced to defend the use of a complementary therapy and belittle the role of science in guiding us to the best therapy. On the face of it, it sounds utterly convincing. Under certain circumstances, even hard-nosed scientists would probably agree.

But what if the treatment in question was tested in a clinical trial and was shown to be completely ineffective? In such cases, there can be a glaring gap between a patient's positive experience and the negative view of science, and in any discussion about complementary medicine, it is essential to understand why this gap so often emerges.

Take Rescue Remedy for example, a version of Bach Flower remedies which is promoted as an anti-stress agent. The best evidence available to date shows that this treatment does not reduce anxiety. That is to say, in the most rigorous clinical trials comparing the effects of Rescue Remedy to those of a placebo, it proved to be no different from the placebo. The remarkable thing, however, is that both real and placebo treatments had some success - a percentage of people improved no matter which option they received.

This success, one has to conclude, is due to the placebo effect which comes automatically as a "free bonus" with any medical treatment. The causes of a placebo response relate to our own expectations which, in turn, can be influenced by a range of factors - everything from time spent with a therapist to money spent on the remedy.

Having understood this fundamental phenomenon, namely that even "ineffective" treatments can help patients through a placebo effect, we can now make sense of the gap between personal experience and scientific evidence. It is possible that the two are diametrically opposite. Mr Smith may experience less anxiety after taking Rescue Remedy, even though clinical trials show absolutely no benefit beyond a placebo effect.

In complementary medicine, enthusiasts often take this one step too far. They extrapolate from an individual experience and turn it into a general rule. The fact that Mr Smith feels better after using a complementary therapy, they say, is sufficient justification for using it on a large scale. The question of whether it helps patients through a placebo effect is, in their view, quite irrelevant. It reduces the suffering of patients, and that is all that really matters.

This argument may seem fairly convincing, but what would happen if the same standards were applied across the board? Imagine a pharmaceutical company marketing a new drug. Tests show that it doesn't work; ie, it is not superior to a dummy pill. But the company could argue that it nevertheless deserves a licence because, as with Rescue Remedy, some people experience benefit after taking it.

The fact that placebo effects depend on people's expectations renders them highly unpredictable. Not only could Mr Smith experience benefit and Mrs Jones not, but as a population, our collective expectation might change unexpectedly. If we hear, for instance, that Madonna, Cherie Blair and the Osbournes all swear by Rescue Remedy, we attach high hopes to this treatment ourselves. If, the next day, we hear that Rescue Remedy doesn't work or perhaps even causes harm, our expectations would probably dwindle. The point is that placebo effects are far too unpredictable to be relied upon.

And what about the ethics of using a treatment which exclusively works through a placebo response? Would it not be misleading and therefore wrong to give patients medicines which are pure placebos and not tell them so? Misleading patients in this way is clearly not ethical.

In my view, the placebo effect is not a sufficient justification for using or promoting a treatment that has no benefit beyond a placebo response. Therapies used in routine healthcare should have both placebo and specific effects - which is the view of the National Institute for Clinical Excellence (Nice) and every other official body in healthcare. A strategy of employing pure placebo treatments would lead us straight back to the dark ages of medicine. Adopting it in complementary medicine is, I fear, ill-advised and counterproductive.

· Edzard Ernst is professor of complementary medicine at the Peninsula medical school, at the Universities of Exeter and Plymouth.