I doubt that you can get 20 colors that are going to look distinct enough from one another to be obviously different (especially if they are filling small areas) in the eyes of the average viewer. You may have to use texture differences as well.

Looking “good together” is highly subjective. Here is my offering to you.

1. Are you wanting to excite the user? If so, use Complimentary Colors.

2. Are you wanting to have a business and clean serious tone? Use A-Chromatic Schemes.

3. Are you wanting to have business, serious tone, cleanliness yet break away from traditional? Select one color, and surround it with a gray scale.

This is my offering to you. Eat well.

This sounds like a bad idea… there’s a reason your palettes are limited to a small number of colours. What kind of information are you trying to represent? Perhaps there’s a better way to do it.

@glacial the totals from 20 different states for each month; not sure how else to do it.

h e r e

Add patterns to the basic 5 colors you choose. for example: checks; polka dots; plaid; vertical lines; horizontal lines; stripes on a slant; hollow circles; plus signs; figure 8’s etc. You can add white and black to your colors too. A combination of these will get you to 20 easily.

Texture differences: ^^^ first post.

Do you have a color wheel?

Look up a color chart online for an artist’s paint manufacturer such as Winsor Newton watercolor paints. That will give you a vast palette to select from but I’m not sure you’ll be able to reproduce them all graphically.

The texture/pattern idea is a good one, but isn’t possible with the javascript library I’m using.

My advice would be to stay away from colour differences altogether. You are not going to be able to choose 20 colours that are different enough from each other for the reader to be able to distinguish them in any meaningful way – which means that colour is useless for your purpose.

If you want to use a bar graph, perhaps do something like the “stacked bar chart” on this page – where you are using only one colour, and the gradation indicates smaller and smaller values. In addition to that, add text onto each section that tells the reader what each one is. So if each section is a state, use the 2-letter abbreviation for the state. Even the pie chart on that page is better than a multicolour version of the same thing – your readers are going to try to infer differences from the colours, and they won’t be able to figure it out because at least two of the colours will be impossible to tell apart. It hurts the brain.

The point is, the reader simply can’t read anything from colour differences in the kind of chart you want to make. So don’t try to ask them to.

Colors from the same family or color can look vastly different. Red, Pink, Cherise, Burgandy and cherry are quite different. Then either cooling a colour or warming it can change it a lot. Consider gold mustard and pale yellow. Same family quite different. This is done by adding grey or reds.

@Shippy This is true, but the reader takes a message away from different values of the same colour – they associate these categories more closely in their minds, and assign a greater or lesser value to the categories depending on the colour value.

There’s a good book on what messages readers take from graphics, called How to Lie with Maps. People who publish statistics in graphical form can greatly mislead their audience, either intentionally or just by not understanding how we read colours and shapes.

If you are depending on differences in color to convey meaning, better not assume that everyone can make out those differences. Some people don’t have the perception for it—any shade of red is just “red” or “pink”—and some simply don’t have the vision. Not to mention the colorblind. I recall this as a huge issue in some wiring diagrams in a technical manual: there were so many lines, and so fine, that you simply couldn’t get enough of any color in one spot to make a clear impression, even before problems of distinction kicked in.

I’m wondering if bar graph and pie chart are, after all, the best way to show your data. Don’t they work best for large chunks of data rather than for fine distinctions?