Your Friend, the Histogram
One of the first revelations that new digital shooters have is that they can see their images as soon as they've been shot. Quite remarkable if you're used to shooting film and not really knowing what you've got until your slides come back from the lab. The second realization though is that it isn't practical to double check your exposure this way, especially under bright, outdoor light. Revelation number three is that they're supposed to use the histogram for checking exposure, but revelation number four is that they have no idea how this is done.
Or for that matter, what a histogram really is.
Histograms weren't invented by camera manufactures (or Adobe for use in Photoshop either). In fact, they come from the field of statistics where they are used to graphically show the distribution of values within a set of data points. The graph has two axes: across the bottom (the x axis) go all the distinct values your data can have, and up the side (the y axis) you plot how many times each value occurs.
In a digital camera, values in 8-bit mode range from 0 through 255. The subject brightness range we can capture in any given image and still retain detail is about five stops. We can therefore divide a scale of image brightness into fifths and consider each one to be about a stop. These divisions can be referred to as very dark, dark, medium, bright, or very bright. This becomes the x axis of our histogram. On the left hand end is black (zero) and on the right is white (255), with every other possible brightness spread in between. Eighteen percent medium gray would be dead center. If we count how many pixels within our image have each distinct brightness value and plot this on the y axis, we get our histogram. Shown here is a sample histogram, but they won't all look this way, nor should they.
So, what is a histogram supposed to look like?
Well, the short answer is that it's supposed to look like it's supposed to look. What this is depends on what we want our images to look like. Contrary to popular belief, an ideal histogram isn't necessarily clustered around the center of the brightness range like some sort of idealistic bell curve. There is no single way a histogram should look any more than there is a single way every image should look. A dark, low-key image will have a histogram that clusters mainly around the left end. The histogram of a bright, high-key image will tend towards the right. One from an image with a more "average" character will likely be spread more evenly across the spectrum. Shown here are two examples with their corresponding histograms that illustrate this point. Looking at a histogram can tell us a lot, but primarily only in the context of how we want the image to look.
We can and should use the histogram to gauge exposure. If the graph is pegged against the far left edge, at least some of our image is completely black, lacking any detail. This may be bad; it may be good. It depends on what we want. Deep shadows and silhouettes are supposed to come out this way. If the histogram is pegged against the right end, we have completely lost detail in at least some portion of the image that will show as pure white. While this may be good, it most often is not. Unlike on the shadow end, burned out white is rarely desirable. Here are three different exposures of the same image and their corresponding histograms. The underexposed one is starting to clip on the shadow end which may or may not be an issue. Given the subject matter though, not having any pixels over medium would definitely be considered a problem. The histogram for correctly exposed one shows a subject clustered near medium, with no completely black or white pixels. The overexposed version clearly shows clipping in the highlights. We have lost data here that can not be recovered later since we never captured it in the first place when the image was shot.
When in doubt, it can be worthwhile to slightly overexpose images, so long as the histogram doesn't end up being clipped on the right. It is far easier to tone down the exposure later in Photoshop than it is to take an underexposed image and try to boost the exposure later. Increasing exposure post-capture will also amplify any noise in the shadows, something best avoided. This technique is commonly referred to as "exposing to the right" and can work on all types of images, providing we adjust the exposure after the fact after uploading these images to our computer.
Don't overdo this technique though. Some cameras like the Nikon D100 show only a single histogram channel rather than separate red, green and blue histograms. What the camera shows instead is a composite histogram, averaging the values from the individual channels. In the case of the D100, this is weighted heavily on the green channel. If an image tends to have a variety of colors, this averaged histogram should be fine, but if it biased heavily towards a single color, we may need to be careful. If we push the exposure too far to the right, we risk burning out a single channel without being able to tell until it's too late. On the D100, the worst case scenario is a large, overpoweringly red subject. With its green bias, the camera histogram can seem off by as much as a stop in such an extreme case. It isn't really wrong of course, but it isn't seeing the image the same way peopel do. Most images are no problem, but watch out if your metered exposure differs radically from the histogram exposure for strongly single color subjects.
Hopefully this will give you some understanding of how to read the histogram for an image. Clearly the best way to learn what a histogram can tell you though is to go out and shoot some images of your own, paying attention to the histograms after each shot.