Tufte begins with the different kinds of informational graphics (maps, time-series, narratives, and relational graphics), describing their origins and evolution and presenting examples of excellence in their design. Many of these are fascinating in their own right — two that I particularly appreciated were Minard's depiction of Napoleon's disastrous retreat from Moscow and an 11th century map of China.
"For many people the first word that comes to mind when they think about statistical charts is 'lie.'" Tufte gives examples of different kinds of deceit in graphics, along with some principles for maintaining graphical integrity. He goes on to consider the reasons for the poor quality of many informational graphics: one is the relegation of their design to those with art training but without an understanding of either the substance of the material or of quantitative (statistical) methods.
Part two begins by introducing some terminology and theory for describing graphics. The principle "Above all else show the data" is formalised as maximization of the data-ink ratio, and illustrated with some "before and after" examples of erasure of redundant or non-data-ink. Tufte excoriates various kinds of "chartjunk": moire vibration (the disconcerting effect caused by repeating patterns), the overuse of grids, and the "ducks" created when the design takes precedence over everything else.
Tufte gives specific suggestions for the design of box plots, bar charts, and scattergraphs. He argues for the use of multifunctioning graphical elements — building data measures or grids out of the data itself, for example, by using labels that also show the end points of the data ranges. And he looks at ways of maximizing data density (within reason) and using "small multiples", or repeated smaller graphics. A final chapter steps back to consider the balance between text, text-tables, tables, semi-graphics, and graphics — "Given their low data-density and failure to order numbers along a visual dimension, pie charts should never be used" — and to touch on the aesthetics of proportion and scale.
All of this is liberally illustrated with examples, drawn from across the natural and social sciences. Despite the space devoted to these, The Visual Display of Quantitative Information packs a lot in, avoiding repetition or verbosity. Tufte's own tables and graphs are appropriately effective and the volume as a whole is elegantly put together: though it's more than that, it could be appreciated simply as a work of art. Tufte also finds room to survey publication practices across a select sample of international newspapers and journals, comparing the data density of graphics and the proportion of relational graphics (involving at least two variables that aren't temporal or spatial).
Most obviously, The Visual Display of Quantitative Information should be read by those involved in writing, editing, or designing documents or displays that contain statistical graphics — from professional editors, technical writers, academics, and journalists right down to high school students. But others may appreciate it too: it has changed the way I look at informational graphics.