Authors need to pay attention to figures for both scientific and technical reasons. Figures are often the primary presentation of one's results, and they need to be clear and understandable. Figures are also expensive to produce, and a bit of care in their preparation will keep journal costs down and production on schedule. The following are just a few thoughts from experience and talking to production staff. I'd welcome any suggestions for modification or additions.

In the old days, we could rely on human artists to make the figures right. Unfortunately, neither computers nor most astronomers have good artistic judgment. The only substitute is careful thought and examination of the figures. A variety of general suggestions can be found at http://www.journals.uchicago.edu/ApJ/instruct.html#sc2 and specific suggestions for electronic art at http://www.journals.uchicago.edu/ApJ/eartguide.html. Additional suggestions can be found below.

The first thing to keep in mind is that most figures will be reduced in size for publication. The final width will normally be one of three values: 88 mm for a single column, 185 mm for full page, and 248 mm for sideways on the page. The single column width is preferred if all the information can be presented at that scale.

Please be sure to examine all your figures at their expected final scale. Borders, tick marks, lines, and labels should be large enough to be clear at this scale. No feature should be thinner than the finest lines commonly used in the Journal. If features are too thin, small bits of metal type break off during printing. The printer does not like this!

Specific suggestions:

  1. Use lines that are thick enough to show up well, especially if there will be a large reduction factor. Consider relative thicknesses as well, making the most important lines (often the figure borders) the thickest.
  2. Consider the relative sizes of all labels. The most important should be the largest or at least no smaller than less important ones. Labels on different axes should be consistent in size and style. It's best to avoid a mix of many different label sizes in one figure; one or two sizes usually suffice. Confirm that all labels are big enough to read! Bold fonts often work better than non-bold, especially at large reductions.
  3. Avoid clutter. Don't put too many tick marks on the axes, and don't label too many of the tick marks. Often two or three tick labels on each axis suffice.
  4. If different symbols are plotted, make sure they are distinguishable. If a grey scale is overplotted with symbols, make sure the symbols contrast enough to be seen.
  5. Avoid wasting space. Don't repeat axis labels unnecessarily. In figures consisting of several panels with the same axes, consider labeling only one of the panels and moving all the others closer together or even juxtaposing them. If panels have one common axis, consider juxtaposing panels on that axis, again eliminating redundant labels.
  6. Consider whether it's better to explain symbols in the figure caption or symbolically with a legend in the figure itself. I don't know any general rule; just decide which is clearer in each case. If you do use a legend, make sure it will not be confused with the data.
  7. There should be no internal grid lines in ordinary two-dimensional graphs. (Possibly they might be needed in some very rare case, but I have yet to see an example.) Grid lines might be needed when attempting to show three-dimensional objects, but alternatives such as an "illumination direction" may work better. The most important thing is to be sure the meaning is clear.
  8. For grey scale (or color) images, a rendering that looks good on a computer screen may not be best for printing on paper. Printed images generally are clearer when there is more white than black. For astronomical images, this usually implies a negative. Careful attention should be paid both to the zero level (which should usually be near white) and to the contrast stretch (which should be chosen to show the key features clearly). The (expected to be) final version should be printed on a high quality printer and examined on paper. For color figures, see the special information.
  9. For figures that represent only a single dimension of data in each pixel (whether the dimension be surface brightness or something more complex like a ratio or other function), I strongly recommend using a grey scale and not a color image. The major reason is that no color palette has an intuitive translation to a one-dimensional scale. Also, color images are expensive, and a non-trivial fraction of astronomers are color blind and may be unable to interpret a color image.
    Color is justified when it is the best way to present a complex data set. Good examples might be presenting images at three different wavelengths encoded in a single figure as blue, green, and red or superposing a red-blue color palette to indicate radial velocity on a surface brightness image.
    If you still want to use color figures, be sure to see the special information.