In 1865, Claude Bernard wrote about medicine as an emerging experimental discipline. This prescient work warned against intellectual traps into which we still stumble today. He noted how easy it is to give greater credence to evidence that supports your views rather than challenges them. He cautioned that experiments should never aim to prove something; rather, their goal should be to test something (lest a desired outcome skew the whole process). He also stressed that experimental results are limited to the experimental conditions — one must be cautious in generalizing to other circumstances or populations.

It's disappointing, but not surprising, that even eminent scientists stray from these precepts. "Chocolate milk in the schools and other products of expert opinion" (June 22) described the path to a now-challenged belief that a low-fat diet is of great preventive-health value.

But I fear that the author of the article — by suggesting that a high-fat diet would be fine — has made some of the same mistakes.

A large recent study failed to show a reduction in cardiovascular mortality through a low-fat diet in adults who were not at high cardiovascular risk. That doesn't really tell us anything about the risk of a high-fat diet, the utility of fat restriction in patients at high risk, the additive value of fat restriction when other strategies are in use, or the importance of dietary fat in childhood. In fact, there have been studies in which dietary fat reductions led to reductions in vascular morbidity when the group studied was a population with high fat intake and high cardiovascular disease incidence (North Karelia, Finland, for example).

The best response to the recent study may be: "Gee; whenever we look closer, this story gets more complicated and it gets harder to decide the strategy to adopt or recommend." A reasonable working conclusion might be that a modest fat intake is a good idea, but there seems to be little gain for most people in getting the fat intake really low.

The last 60 years provide helpful context. Long ago, cholesterol deposits were found to be important in hardening of the arteries, and the blood cholesterol level was a risk predictor. One could produce hardening of the arteries by feeding cholesterol to animals. The answer seemed to be "eat less cholesterol." But purity of the cholesterol made a big difference in those experiments, and most of the cholesterol in the arteries had been made by the patient or animal (it wasn't from diet).

That led in a new direction: limit cholesterol production. Drugs known as "statins" helped achieve that goal, often combined with a low-fat, low-saturated-fat diet. An additional wrinkle has recently emerged, in that statins have benefit even when they don't lower cholesterol much; they have some anti-inflammatory effect that could be a surprisingly large part of the story. (There are additional risk factors to consider, such as diabetes, hypertension and smoking.)

Against a background this complex, failure to demonstrate a benefit by manipulating one factor does not tell us the factor is unimportant — it just sets rough limits on how important it can be in the population studied and in the range that was scrutinized.

So where does that leave us? We don't know that it's OK for children to eat high-fat diets. We also don't know if there is benefit from encouraging them to eat low-fat diets. Did the recent study make us much smarter about these questions? Unfortunately, no.

In deciding policy about school lunches, we must admit the limitations of our knowledge and be prepared to modify our recommendations as we learn; it's also important that the public understand science as a process of successive approximations of the truth, with false starts and occasional complete surprises.

And while we worry about diets, let's remember the modifiable risks that have withstood scrutiny better than dietary manipulation of fat. If we can get kids to exercise regularly and not start smoking, the benefit will probably dwarf that of anything we do with diet.

Dale Hammerschmidt, of Minneapolis, is emeritus professor of medicine at the University of Minnesota.