The "Flynn effect" is the increase in intelligence over the past 100 years. Populations 100 years ago, in fact, had IQs of about 70. While not all of this differential is due to a "real" increase in intelligence (a substantial fraction of it probably relates to test-wiseness), some of it probably is. However, "general intelligence", the biological substrata which cause the positive manifold amongst different IQ tests, has not increased, since the sizes of Flynn effects on different tests are inversely related to those tests' g loadings. The same pattern holds amongst items. Thus, for example, vocabulary size (generally the most or among the most g loaded tests) has shown a small Flynn effect, and by some measures even a decline. However, the "Coding" test (from the Wechsler) or the "Draw-a-Man" test both have low g loadings and have shown very large Flynn effects. Nevertheless, some reasonably highly g loaded tests, such as the Raven's Progressive Matrices, have shown significant Flynn effects. It is proposed that this large Flynn effect relates to a property of tests called "rule-dependence". Tests vary in the degree to which "rules", patterns, or heuristics can be applied to different items, or in the degree to which solutions to one item are relevant to the solutions of other items. For example, general knowledge is not rule dependent at all. The Classifications subtest from the CCF is moderately rule dependent, since there are a few principles on which the shapes in question are classified (size, orientation, number, etc.), which must be learned and reapplied. (Spearman [among, I think, other researchers] listed some "principles" of verbal analogies or classifications, such as similarity in function, appearance, taxonomic category, etc., which are, of course, related to the above description of the Classifications subtest.) The Raven's is the most rule dependent test of all, since it requires only a few cognitive operations or principles (addition of shapes, number of shapes, distribution & superimposition of shapes, etc.), which are reapplied to all items. It is proposed that the Flynn effect sizes on different tests relate to the rule dependence of different tests: perhaps because cultural exposure to relevant principles acts more on more rule-dependent tests, perhaps because people's ability to infer rules has improved, perhaps because they can better store rules in working memory for future application. Finally, the long-term causes of improvements in rule-dependent tests are discussed. Improved education and slowing life history are plausible factors, but low-IQ nations (which tend to be poorly educated and fast LH) do not score any lower on the Raven's than on other IQ tests, suggesting that current international differences in education and LH do not depress scores on rule-dependent tests any more than on other tests.