(410k words / 840 pages; online edition; WP) One of the classics in the field, Jensen sets out to explain almost everything, it seems, in psychometrics, from the core concept of error-prone measurements and extracting factors to the various tests available, their correlates, concrete justifications for why the normal distribution is more than an assumption of convenience (a number of the points were new to me), exhaustive coverage of the core topic of various kinds of bias and evidence against them, to culture-fair tests, and finally how mental testing is best employed. (There is also some discussion of behavioral genetics and what the genetic architecture of intelligence might be, but that's a minor topic and he gives more attention to other things like reaction-time research.)

Discussion of the topics straddles that fine line between too informal and too formal, as Jensen is careful to introduce and explain each concept as he goes and includes excellent summaries at the end of each chapter to the point where this would make a good textbook and it is so readable that I think even new tudents to statistics could understand almost everything in the book (at least, as long as they paid attention and occasionally checked back to the glossary to be reminded of which of the many formulas is relevant to a particular point; there is a ton of content and skimming will not work).

Overall, my impression is extremely positive. I'm especially impressed that despite now being 35+ years old (and hence based on research from before then), there's hardly anything substantive I can object to. The statistical principles are largely the same, the black-white gap has hardly budged, the lack of bias remains accepted, etc. I saw no large mistakes or content that has been totally obsoleted, and in some areas one would have to say Jensen is being constantly vindicated by the latest research - in particular, in arguing for the genetics of people of non-retarded intelligence being largely uniform over the intelligence range and governed by a large number of additive alleles (yielding an objective normal distribution), none of it needs any correction. Afterwards I read a recent review, "Bias in mental testing since Bias in Mental Testing", Brown et al 1999, comes to the same conclusion.