BALTIMORE -- Supat Sirivicha has a way with pizza sauce. He mixes it with petroleum ether, a flammable liquid, and watches the sauce separate into layers like Jell-O 1-2-3. Sirivicha is not having the ladies over for jello mold and canasta, however. Nor is he trying to set a pizza parlor on fire.

As laboratory director of Strasburger & Siegel Inc., a company that analyzes the nutritional content of foods, Sirivicha has just completed one of the first steps in analyzing the fat composition of pizza sauce. Fat, which dissolves in ether, floats to the top, and heavier, water-soluble components -- such as starches and sugars -- sink to the bottom.

The fat can be skimmed off and reduced by a computer to peaks on a line graph. After more than four hours of chemical analysis the company that makes the sauce will know how much saturated, mono-unsaturated and polyunsaturated fat is in its product.

With the Food and Drug Administration's proposed requirements that the labels of all packaged foods carry nutrition information -- including the amount of saturated fat -- private labs such as Strasburger & Siegel are gearing up. Already, business has "increased tremendously," according to Alan Parker, president of the Baltimore firm, whose clients include the Campbell Soup Co., ConAgra and Nabisco Brands Inc.

Yet, while nutrition-conscious shoppers may eye the list of calories, fat or fiber on food labels as if it was etched in stone, food analysis is far from an exact science. The method of testing, the sampling procedure and natural variations in foods can all lead to human and mechanical errors. Plus, the information on the label represents only what's in the package; once consumers get it home from the supermarket and cook it, its nutrition profile may change dramatically.

"When you put a nutrition label on a packaged food, you're trying to give the consumer a basis for comparison and that's all, said Phil Oles, manager for the food groups at Lancaster Labs, a private lab in Lancaster, Pa. "You're not really trying to tell them 'you're going to ingest two grams of fat when you eat this.'"

Maggie Glavin, deputy administrator for regulatory programs at the U.S. Department of Agriculture's Food Safety and Inspection Service, said, "We have done a very bad job" of showing the public why labels aren't exact. Shoppers think "if a package says '120 calories,' there are not 119 or 121. That's simply not true," she said.

The flip side, Glavin continued, is that "it doesn't really matter if it has 119 or 121 calories in it. Your world isn't going to end."

Of course, it's all a matter of magnitude. In May, the Nevada firm that made Phoenix Fiber Cookies recalled the product after the FDA notified the company that its promotional materials violated federal laws. The cookies were promoted as useful for weight control and reducing the risk of cardiovascular disease, colon cancer, diabetes and diverticulitis. One of the violations included claims that each cookie contained 120 calories; FDA tests showed that they contained twice that amount.

Government regulations do allow for a margin of error for nutrition information, but it doesn't go that far. For calories, fat and sodium, foods can contain up to 20 percent more than the declared value on the packages. For example, if a product is labeled as having 10 grams of fat per serving, it could have up to 12 grams and still be considered properly labeled.

For vitamins and minerals that are added by the manufacturer, the declared values must be exact, but for those that occur naturally in foods, the product must contain at least 80 percent of the stated amount. For example, if a package says the food contains 10 percent of the recommended daily allowance for Vitamin C, a minimum of 8 percent would be permitted.

There are many reasons for these variations. For one, some scientific tests are more accurate than others, as are the labs that perform them.

The method used to determine the amount of fiber in a food, for example, "is the least well established," said F. Edward Scarbrough, acting director of the FDA's Office of Nutrition and Food Science. "It's still somewhat controversial."

As for extracting and analyzing the amount of cholesterol in a food, the current methods are imprecise, according to Oles. Lancaster Labs has developed a method that Oles believes is "much more robust."

Aside from variations in scientific methods, the number of samples gathered can affect the outcome of the analysis. Generally, the more samples that are analyzed, the more statistically accurate the results.

In July, amid much fanfare, several fast food chains announced that they were switching from beef tallow blends to all-vegetable oils to fry their french fries. The switch reduced the amount of saturated fat in an order of fries, but because all oils have the same amount of calories and fat, the calorie and fat counts shouldn't change.

Nevertheless, Wendy's International's six-ounce serving of french fries fried in a beef tallow blend went from 550 calories to 450 calories fried in corn oil, and the fat content dropped from 29 to 22 grams. After the oil change, the calories and fat in McDonald's and Burger King's fries remained the same.

Denny Lynch, spokesman for Wendy's International, said the discrepancy occured because the sample sizes were different. French fries fried in the beef tallow blend were only analyzed from one location; samples of corn oil fries were pulled from numerous stores all over the country.

Yet even though the data from the corn-oil fries represented a much larger sample, "if we took the same stores and did it today, we wouldn't get the same numbers. This is not an exact science," Lynch said.

Wendy's doesn't always analyze products as they are sold in its stores. Sometimes, the fast food chain sends uncooked samples to a lab, which cooks the food and analyzes it, or the company cooks the food in its own labs, packs it in dry ice and then sends it off to the lab.

McDonald's previously sampled food directly from its restaurants, but switched to sampling food from its processing plants because of the sheer number of locations. "I have 9,000 sites in the United States alone, said Michael Goldblatt, director of nutrition and product development for McDonald's. "I have 60 crew people who have been certified to prepare a product. That alone is 540,000 manufacturing units."

While technicians at the plants compose the ideal hamburger, Goldblatt acknowledged that food in the restaurants may be more variable. When it comes to french fries, for example, some employees "like to stuff" a package, while others understuff, he said. Nevertheless, the chain does analyze food from individual stores as well, and Goldblatt said there has been very little variation in the nutritional content from food sampled in the plants. "I did what I thought was reasonable," said Goldblatt, who designed the sampling program for the chain's nutrition brochure. "I have no idea if it's right."

For processed food products, the standard procedure is to analyze a composite made from 12 different samples of the food, which are whirred in a blender, grinder or food processor to get a representative sample. FDA procedures state that the 12 samples should come from one production lot. (USDA, which regulates food labeling on all meat and poultry products, has more stringent guidelines. Food must be sampled from 12 consecutive lots.)

With either method, there may be differences in one production lot (say the raisin dispenser doesn't uniformly drop the same number of raisins into each cereal box), between production lots, as well as differences in plants and plant locations (corn, for example, differs in its nutritional content depending on where it's grown).

As a result, some firms may analyze far more samples, according to Janet Dudek, director of the chemistry division of the National Food Processors Association. "Some companies have been sampling for years and years and have hundreds of numbers," she said. "Other companies may just have a few."

Not only are there natural variations in foods, but the nature of the food may also make the analysis more or less precise.

For instance, the nutritional content of pepperoni pizzas is a lot more variable than solid chocolate candy bars, according to Oles. "A pure chocolate bar is going to be completely homogenous. Every piece of the chocolate bar will be the same."

Another factor that can influence the outcome of the analysis is the state the food is in when it's analyzed. While most food manufacturers try to have their foods analyzed as close to the form in which they are eaten, there may be differences of opinion in how the food is eaten.

In the case of tuna fish packed in water or oil, "do you make the composite on the tuna only or the entire contents?"asked Parker. For certain nutrients, it can make a big difference. Sodium values for tuna packed in water may change dramatically, for instance, depending on whether the product is analyzed with the packing liquid or without it. Parker noted that Strasburger & Siegel has analyzed tuna "every which way."

Dudek said that most canned fruits and vegetables are analyzed with their packing liquids. "The assumption is that people will eat it {the packing liquid}," Dudek said.

For those who don't eat the packing liquid, the difference in vitamins and minerals could be substantial, however. Dudek said that after processing in the plant and cooking in the home, 40 to 50 percent of some water-soluble nutrients, such as B vitamins, can migrate into the packing liquid.

Additionally, the quantity of nutrients that leaches into the packing liquid depends on the cooking time. "I may take a can of peas and warm them for two minutes. I may also be the consumer who walks away and eats them after they've been heated for 10 minutes," Dudek said. The nutrient profiles between the two portions of peas, she said, "will be very different."