The introduction immediately identifies the purpose of the survey as being about rape and sexual assault, announcing immediate intentions encourages more reporting, whereas NCVS begins by asking about crime in general.
The questions are all phrased "how many times has x happened" presupposing x has already happened, which brings up false positives. It's well established that wording can widely vary answers recieved.
The question on drunken sex is so poorly constructed , and accounts for over two thirds of the positives.
And is near the end. Specifically the respondent is not instructed not to list episodes of drunken sex that were consensual.
And it's not minutia we're talking about, BJS comes up with just over a quarter million total rapes for 2010, and CDC insists there were over 3 million that year. Too much a disparity, and the BJS in general is more highly regarded on numbers anyway.
The CDC has a problem in that they conduct studies meant to be used for political purposes