It's official! Coffee-drinkers are less likely to suffer heart disease than tea-drinkers. A recent study by scientists at Dundee University's medical school, published in the Journal of Epidemiology and Community Health, shows that the higher the coffee consumption, the lower the risk of heart disease and death. The researchers studied 11,000 men and women aged between 40 and 59 and found that the risk of dying from heart attack doubled for men if they drank five cups of tea rather than five cups of coffee. Women who drank tea were 50 per cent more likely to suffer a fatal heart attack than those who drank coffee. So the message is clear: avoid tea, drink coffee.

In fact the message is about as clear as a frothy cappuccino. In July we were told by the London-based International Anti-oxidant Research Centre that tea actually reduces heart attacks because it contains flavinoids, protective chemicals that fight free radicals and prevent arteries from clogging up. But the contradictory messages from the laboratory are not limited to tea and coffee. For years scientists insisted that sunbathing was harmful; then new research claimed that sun was good for us. Wine, particularly red wine, was supposed to be good for us in moderation. Then a recent study of Scottish men showed that heavy wine-drinkers were twice as likely to die from strokes as non-drinkers. Similarly, if you eat red meat in the United States, the probability is high that you will end up with bowel cancer. But if you eat red meat in England, no such probability exists. If you are a vegetarian, then peanuts are an excellent secondary source of protein; if you are a woman, then they would reduce your chances of dying from heart disease. In either case, the probability that you will acquire the potentially lethal peanut allergy, and die on the spot, is far from negligible.

So what are we to believe? The first thing to realise with claims such as "X is good or bad for you" is that they emerge largely from the domains of what I call flaky science. Flaky science usually deals with such subjects as health, nutrition and safety. The practitioners of flaky science are not bad scientists; they are just naive victims of their own training. I suppose it wouldn't be so bad if the public got a word of apology every now and then for being misled. But I've never heard one of these scientists say "sorry". The Pope himself would be more willing to admit fallibility than the men in the white coats who tell us what is good (or bad) this year.

The key to the illusion of the flaky sciences is the uncritical application of statistical methods. For statistical tests to have any validity, they must be framed quite precisely around a well-defined hypothesis and subject to strict protocols of collection and treatment of data. Otherwise they prove nothing more than you could do by finding interesting numbers in the phone book. This works well in fields where the real questions can be formulated precisely. But for complex subjects such as health and well-being it's like using a scalpel to tidy up a garden: the tool is very efficient but totally inappropriate. The material is too variable, and the concepts too fuzzy, for such precision to be meaningful. That's why a slight difference in test design can make all the difference to the answer, and you can prove anything you set out to prove. Hence so many contradictory answers from research based on statistical studies.

Everyone knows that people manage to lie with statistics, but it is not always clear how scientists so easily fool themselves along with their audiences. We can identify flaky science by asking a few simple questions.

First of all: who is funding the research? If the research that "proves" that wine is good for you is funded by the wine industry, then we know where we stand. The possibility of conflict of interest cannot be ruled out. The second question relates to the size and thoroughness of the study. A "longitudinal" study over a period of years, with thousands of subjects, is in a different league from a quick survey on a particular local group.

Third, we must check whether the results are stated with "confidence limits". No statistical test, if honestly reported, will say "yes" or "no". Instead the conclusion is given in terms of the rough betting odds that it is correct. Such are confidence limits; 95 per cent means that it's probably OK at 20-1. If a conclusion has no confidence limit, then it doesn't deserve much confidence from its readers.

And then, if you really want to see what a test is worth, look at the sample being tested. Is it representative for the research question? Age, gender, income and lifestyle can all influence people's answers, and if these are not taken into account, bizarre results can come out. Statisticians speak of "confounding variables", which are the real determinant of the answer. We know this from the school league tables: it's easy to get good exam results with affluent children. If social class is allowed to "confound" such results, then, while thinking we are measuring quality of teaching, we are only measuring parental income.

The "coffee is good for you" study provides us with an excellent example of these pitfalls. It turns out that the "somewhat surprising" result is due to coffee-drinkers being younger and richer. Tea-drinkers, on the other hand, tend to be older and associated with increasing social and economic deprivation. So what the study actually confirmed is that the poor suffer from more ill health, not that coffee is good for you. Professor Hugh Tunstall-Pedoe, the team leader, knows well that his study does not prove anything about the beneficial effects of coffee. "We are not claiming coffee is beneficial and tea is harmful," he has declared. So just what is the study claiming? Simply that the "marvellous preventative" qualities of tea are "incompatible with our result". Which is not quite the same thing as saying coffee will save you from an early heart attack.

While Tunstall-Pedoe has been throwing scorn at tea, a group of Dutch scientists has been claiming that chocolate can protect people against cancer and heart disease because it contains the same antioxidant properties as tea. Research published recently in the medical journal the Lancet, based on a sample of 6,250 men and women aged between one (yep! That's what they say) and 97, suggests that chocolate contains high concentrations of antioxidant compounds called catechins. These are believed to protect against cardiovascular diseases and possible cancers.

So drink your caffeine, gorge yourself on chocolate and be merry while both still have scientific blessings. No doubt next year they will become scientific, and social, pariahs. Personally, I trust the scientific judgement of Mark Twain, who declared all good things in life to be illegal, immoral or fattening.

Ziauddin Sardar is co-author, with Jerry Ravetz, of "Introducing Mathematics", published by Icon Books at £8.99