The nutritiondata.self.com website that November cited is also the one that I use for my analytical work. I might add that the detailed type of nutrition data as is given at that website is also sometimes available from the companies behind any given product. However, they don't usually make that kind of information public. Sometimes they will provide it upon request but more so for professionals than for ordinary consumers or for guys like me who might be trying to reverse engineer their products. Often that more detailed information will report specific values for nutrients, albeit small in many cases, that show up as zero in the final Nutrition Facts, because of rounding.
As for the matter of rounding and how the FDA requires that nutrients be reported on nutrition labels, the document that I often refer to is the one at http://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfcfr/cfrsearch.cfm?fr=101.9
. For our purposes here, the relevant parts run from "Calories, total" to "Protein". One of the things that you will notice is that the rounding is stricter for certain nutrients than others. For example, the rounding is stricter for "bad" guys like fats (including the current villain Trans Fat), cholesterol and sodium, and less so for "good" guys like dietary fiber and protein. In either case, the swings in possible nutrient values can be quite high. Percentage-wise, they can be more pronounced for small amounts of nutrients than large amounts.
Companies will also play games to keep certain nutrients at zero in their nutrition information. A good current example of this is in the reporting of Trans Fats. Under FDA rules and regulations, so long as trans fats are less than 0.5 grams per serving, it can be reported on labels as 0, with that fact often being trumpeted on the labels and packaging and advertising for all to see. But the 0 number is usually interpreted by consumers as actually being 0. But if more than one serving is consumed, the trans fats number is no longer 0. The entire margarine, margarine-like and shortening industries is being turned on its head because of the Trans Fats rules.
There is also something that is called the 20% rule. What that means is that a reported nutrient value can be off by 20% from its actual value, on either side, and not be subject to scrutiny by the FDA (or USDA). I have read reports that the people who come up with nutrient values often game the system by reporting lower numbers for the "bad" guys and larger numbers for the "good" guys. And they know that they can get away with it because the FDA won't act unless people are getting sick, or false or questionable health or nutrition claims have been made, or their products might be recalled, etc. Also, they know that the FDA does not have the staffing to enforce their own rules and regulations. I have seen cases where I believe even the 20% rule has been violated.
There are also food processors who intentionally overweight their products so as not to lay themselves to claims that they are stiffing consumers by selling them underweight products.
It is also important to consider how nutrition information is created. The two basic methods that I have identified is the use of software or independent test labs that do destructive testing. These days, it seems the more common method is to use software. Some companies, like Papa John's, for example, uses software in its own facilities using its own personnel (or so I was told), whereas others farm the work out to companies to come up with the final nutrition information, usually working off of extensive data bases. Sometimes the specific software used (like Genesis), or the outfit that created the nutrition information, will be stated at their websites where nutrition information or allergy information is given. Using test labs seems to me to be a more accurate way of arriving at nutrient values but that approach seems to be on the wane.
The point of all of the above is that there are enough holes in the nutrition numbers to drive a truck through. In some cases, it might be a small truck but it might also be a big truck.