SKIP Suppose you want to create a document (Web page, e-mail, whatever) that contains some Chinese characters, not pictures of them but real live UniCode characters from the CJK (Han unification) planes? How do you find the appropriate characters to enter into your document? It depends on what you know about each character:

If you already have the desired character in some other document, such as some of your earlier writing, or an online Chinese language "newspaper", and you can find the particular character when you need it, then you can simply copy and paste.
If you know how to pronounce it in Mandarin or Cantonese, and you know how to code that in pinyin, you can use the pinyin to directly key it on the keyboard of your computer, if it's so equipped, else you can use the pinyin to find the character in an online Chinese-language dictionary, then copy+paste from there.
If you know what the character means in English, you can look it up in an online English-to-Chinese dictionary, then copy+paste from there.
But suppose you have none of that information. You've seen the character in visual-only form, such as in a Chinese-language newspaper (hardcopy) you've found lying around, or within open-captions on a Chinese-language program on television, and there's no Chinese-English bilingual person handy to give you the slightest clue what it means, and you don't know enough Mandarin or Catonese to be able to follow the rest of the characters in the audio dialog to be able to hear (by process of elimination) which sound corresponds to that one character that arouses your curiosity? You need some way that you can somehow describe, to the computer, how the charater looks visually, and have the computer determine which character you are describing. That's the topic of the rest below.

SKIP
Methods to look up a character by its shape:

You might be able to use a scanner to produce an image of the text, then use Chinese-language OCR (Optical Character Recognition) to identify each character and produce the sequence of UniCodes for them. I'm not aware of any such software available, and this seems like an awful lot of hassle, especially if you have to travel to a public computer lab to get access to the scanner each time you want a few characters scanned.
If you have easy access to the scanner, but don't have any OCR software, you could scan the text and post to the net, asking if anyone recogizes these characters and could please key them in for you (or use the other methods listed below) and return the result to you online. You could even pay people online to provide this service for you, if you have money, or after NewEco implements third-party contract work. Or you could use the scanned text within a ReCaptcha system.
Sometimes you can identify the primary "radical" within the character, and count the number of additional strokes in the character. If so, you visually look up the radical in a list of standard radicals, then key in the number of additional strokes, then look down the list of results hoping to see the complete character. This can fail because there are two or more similar-looking radicals and you picked the wrong one, or because you miscounted the number of additional strokes, or because you simply didn't recognize the character even though you were looking right at it as you scanned a list of several separate pages of a list of over a hundred characters that share the same radical and same number of additional strokes. My personal experience is that the process seldom worked for me, and even when it did, it was an awful lot of work just to find one character.
There's a system called "four corners" for identifying Chinese characters, where you type four digits somehow identifying what shape is located near each corner of the character, and the lookup engine shows you all known characters that share all four corners. But I've never seen an explanation what numbers to use for what shapes, so I've never been able to assess how well this system actually works.
For common characters, you can eyeball-scan an online newspaper, looking for it to appear "at random". For the very most common characters, you are likely to see each several times on each page of text, allowing you to quickly eliminate your workload to just the few not-so-common characters you are seeking. I found this method much easier to use than the radical+strokes method, getting more than half my desired commonly-seen simple-looking characters completed within a couple hours of searching.
But what if none of those existing methods are suitable for the remaining characters which aren't common enough to find by accident while scanning online newspapers? Below I propose some new methods for directly describing the various strokes of a character into a HTML FORM, whereupon the WebServer application could look for your description within a database that I have set up. (But this won't help me when I'm first setting up the database!! I'll probably use the services of a Chinese-English bilingual person, if I can find any. Such services would not be suitable for "every time I see a character I'm curious about", except to pass time at a senior lunch if I happen to be sitting near a Chinese person, but would be suitable when doing a one-time inclusion of that character into a database.)

SKIP
New ideas for describing a Chinese character:

Use a JavaScript or Java-applet client-side application to track the mouse to directly draw strokes and record the key information about what was drawn. Of course this wouldn't work at all on cell-phones, nor within a text-only browser such as lynx, so I haven't considered implementing it myself.
Update 2013.Jul.17: @rudharcom found a Website which allows me (in Google Chrome for example) to use mouse to draw the character (using JavaScript tracking presumably) then submit it for recognition. For simple charcters it finds the desired character more than half the time, but for very complex characters it almost never works.
Start with blank image, and overlay several possible most-prominent strokes within characters, and have user select from them somehow. Then show the selected most-prominent-stroke overlayed with several possible next-most-prominent strokes, and have user select from them somehow. Keep going in this way adding one new stroke with each user interaction. At any point when the strokes so-far are in fact a completed character, show info about that character before showing the overlay+options to skip that character to build towards a more complicated character. For a partial prototype of this method, see here
Instead of having a limited number of pre-defined strokes to identify visually for selection, have a grid of checkboxes, and check the locations of the start and end and each bend (if any) in a single stroke. Since strokes always start at upper left corner and proceed to lower right, and never complete a closed shape within a single stroke, in most/all cases the computer can unambiguously decide which marked location is the start and the sequence of marked locations from there to the end. Of course this method would work efficiently only on devices that have a mouse. It would be a royal pain on a cell-phone or text-only Web browser to need to step through all the unused checkboxes to check just the few that are being used.
Use a grid, but manually key in the row-column indexes instead of checking boxes. This would be more awkward than checkboxes if a mouse is available, but would work better on cell-phone or text-browser or any other device not having a mouse. To protect against row/col coordiates getting mixed up, an "algebraic" notation could be used, where letters denote columns and numbers denote rows.

Chinese references/tools
.
.
.
.
.
.
.
.
.