Francis Crick's What Mad Pursuit is what I'd like all scientific memoirs to be, or literary memoirs for that matter; it's mostly about the "logic" and the (loose and miscellaneous) methods that working scientists use to do things. As an exposition of the science it isn't particularly good; the main problem, I think, is that there aren't any figures, and without figures a lot of the talk about model-building and crystallography can seem baffling. [I was reminded of Steven Weinberg's general-audience talk at Illinois a couple of years ago: he was trying to explain spontaneous symmetry-breaking without slides, and tried to make up for his inability to conjure the right mental image by gesticulating wildly.]
What Crick does brilliantly is explain why he chose to go into biology (the "gossip test" -- if you find yourself gossiping about something it's probably something you want to work on), how he spent years reconstructing molecules from X-ray diffraction patterns, how the genetic code was decoded, etc. His basic take on Rosalind Franklin is plausible: he felt that she was overly cautious because she wanted to appear "professional," which was natural for a woman in the field at the time, even when a slapdash and unrigorous approach was likely to work better. This still makes her a tragic figure, of course, but for different and more believable reasons than the generic victim story. George Gamow, who briefly worked with Crick on the genetic code, gets a rather appealing cameo, as does William Bragg.
The chapters on the discovery of messenger RNA and the decoding of the genetic code are particularly good. The story with the genetic code goes something like this: Crick was interested in the question of whether bits of RNA -- which is half of the double helix, and has sticky ends, and which Crick assumed to be floppy -- could accidentally get zipped up while wandering through the cell; he figured that if so they'd be unable to make proteins, and certain traits might end up "suppressed." In particular, certain mutants might end up "canceling out" by gluing onto each other. Clearly they would have to be somewhat far apart on the strand of RNA to be likely to do this. However, when this prediction was experimentally tested they found that suppressor mutants -- which it turned out did exist -- were usually near each other, so the original hypothesis wouldn't do. Eventually they figured out that the reason the suppressions happened was as follows: a lot of mutants are deletions, and the genetic code is read three characters at a time. Therefore, a sequence that made sense as:
ALL WET CAT PEE IS# WET
upon a single deletion turns to
ALL WEC ATP EEI S#W ET...
i.e. everything to the right of the deletion is garbled, whereas if you have three deletions nearby you get
ALL WEC APE EIS WET
i.e. the damage is localized, and the original instructions can more or less be deciphered. This was confirmed after a lot of hard work and some clever experiments.
Crick's basic message -- especially to physicists going into biology -- is that theory in biology requires a lot of the nimbleness that this story exemplifies; one has to keep one's perspectives fairly short, and getting too far ahead of the evidence has disastrous consequences. A corollary is that existence proofs, which he calls "don't worry theories," showing that there is an imaginable structure or pathway that does something, count for little in biology.
A final point I found interesting: the question of the uniformity of the genetic code across all sorts of organisms. This is a somewhat puzzling fact, and it seems apparent that there must have been some sort of bottleneck. Crick couldn't think of anything particularly good so he suggested panspermia; Carl Woese and Nigel Goldenfeld (and others, I imagine, but I'm not too familiar with the literature) have since been working on a less unappealing answer, which has to do with the fact that bacteria and archaea reproduce largely by squirting DNA at one another.