For the past twenty years as both a genealogist and librarian, I have been lecturing and writing about the difference between primary and secondary sources when doing genealogical (and for many years legal) research. Now it seems that the Board for Certification of Genealogists has invented a new term: derivative source. I think this muddies the waters tremendously. I've never attended a lecture about this new definition, but have read blog postings regarding it. I understand the terminology, but from a practical point of view, does it matter? The answer is no* (that is no with a caveat).
Let's assume that you are a hard-core genealogist, one who wants more than what the average on-line family tree may offer. You find a birth, marriage or death in the following work: Vital Records of Plymouth, Massachusetts to the year 1850 compiled by Lee D. van Antwerp, edited by Ruth Wilder Sherman (Camden, Me.: Picton Press. 1993). For an example we'll choose my ancestors' marriage record on p. 653: William Renalds and Alis Kitson were married 30 Aug. 1638. This is what I (and all other genealogists up to this point in time) would call a primary source. If you wanted to use this source for entry into the DAR, SAR, Mayflower Society, or any heritage organization, they will accept it. Without question. There is no organization who will send you to find the actual record.
Say you want to publish an article on this couple (like I did) in a scholarly journal like, The New England Historical & Genealogical Register (like I did). This will suffice for a footnote. And it does as footnote #2. [Actually in this case, the marriage appears in two primary sources and I cited them both. However, they would both be called derivative. I've also seen the original record on microfilm.] So, what's the big deal? If the scholarly and heritage-based genealogical communities are OK with this, why invent a new term? Do you really need to go check the original record? No, unless you are completely anal or have reason to believe that the transcription is incorrect. Keeping in mind that any record in print or in a database is "derived" i.e. transcribed by an individual, who in most cases was human and, therefore prone to errors, is part of being a genealogist. I would point out that even many "original" records have been transcribed, and the set we may be looking at may not be the "original" but also a derivative. Just one that is 300 years old.
The one caveat is precisely what I just mentioned. You have an inkling that the record has been mis-transcribed, or there is something missing. For a great example of that see: “Mary Porter, Wife of Uzziel4 Rea of Danvers and Topsfield, Mass.” by David Curtis Dearborn, The American Genealogist 65 (1990):143-7. Indeed Porter should be Potter. However, genealogy is hard enough already without making it more burdensome. Primary sources are sources generated by the parties involved at the time of the events in question: vital records, church records, deeds, probate records, court records, diaries, family Bibles, gravestones, town records, and census records. Secondary sources are those that have synthesized these records into a narrative of some sort. They can be compiled genealogies, town histories, county mug books, biographies, and online family trees. These sources range from excellent to not so much. Unless you've encountered a brick wall or the other available evidence is bringing a source into question, then go for the printed record. You've earned it.