Novices to photography always assume their camera meters magically should always give the correct exposure, but alas, they need to learn that life is not that simple. Meters are pretty dumb. They can be a good guide, but metering is an art done by the photographer.
Here is a quick rose, intended to mimic some real photo scene. Photographed below five times, all with the same settings (automatic point&shoot TTL flash). The ONLY difference in these next pictures is that the background is modified (ordinary white and black craft paper). All else is the same. The camera is aimed at the subject, and TTL automation exposes the picture. However, the camera's reflective light meter sees different scenes differently. Under or over exposure depends on the scene the camera sees. All of this page is speaking only about reflected meters, like in cameras (two more pages to Incident meters).
To make that be real clear, let's adjust the background a little at time, all else stays the same (but still with automatic TTL metering of the flash). At each frame, the ONLY difference was that the black paper was shifted left a little, and then the shutter button was pressed again. The scene the meter sees is different, different ratios of dark and light colors. Then the TTL metering reacts to the new scene it sees. The exposure you get depends on what the meter sees. This is how reflective metering works (it sees the light reflected from the colors in the scene).
This picture of a rose simulates some real picture.
Direct flash, off camera. Automatic TTL metering mode. No cropping, ISO 200, all f/8, 1/200 second.
No adjustments were made, except the black background was simply moved left.
The flash TTL system meters on its fixed central area. Which includes the rose, and some of the background. The meter has no brain, it sees a blob of light with no clue what anything is, or what it means, but it can measure the light. As the background becomes more white, the blob appears brighter, so reflective meters reduce the exposure, still trying to achieve the same overall (middle gray equivalent) average result for the area... not too bright, not too dark.
A light meter does not try to recreate the correct tones. It can't do that, because it only sees a blob of light, and it cannot recognize anything. Metering just tries to keep the average tone more in the middle, not too dark, and not too bright. And this often works out pretty well, for "average" scenes, but there are many exceptions, to which we must pay attention.
The exposure varies with the scene. More black or dark color causes overexposure (trying for a middle tone result). More white or light color causes underexposure (trying for a middle tone result). The meter's goal is to always create a middle tone. The one that is 50/50 actually averages near middle gray, and so comes out about correct. The variations are the expected result. You can count on it. Simply how reflected meters work, all of them.
Reflected meters are aimed at the subject from the camera, and meters the light that the subject's colors reflect.
A white background or subject reflects a lot of light, which reads high, so the meter underexposes the picture.
A black background or subject reflects little light, which reads low, so the meter overexposes the picture.
The expected reflected goal is that all results come out middle gray brightness, not too dark, not too bright. This is all the meter can do (it cannot recognize anything). Fortunately, many typical scenes contain a random mix of dark and light colors that probably will average out about middle gray, then the middle gray result can often be about correct. When otherwise, we can compensate it, to correct it to come out as as we want, brighter or darker, as needed.
Incident meters are the reverse, aimed at the camera from the subject, which reads the incident light from the light source directly (is never influenced by the "subject").
So then any subject tone, be its colors light, dark, or middle, is shown as it is. Kind of a big deal. Point&shoot where it counts. :) Incident meters have the accuracy that newbies imagine their reflected meters ought to have (but can't). Of course this is not possible at the camera, incident meters meter the light incident on the subject (from the subjects position).
Said again, speaking of reflective meters (camera meters), their method is, if more dark area, make it brighter. If more light area, make it darker. This is simply because - its only capability is to make the metered area of all pictures average out to be a middle tone, which is "correct" by its rules, but it may or may not be the result you want. Whatever the scene, the reflective meter's overall goal is that the metered area will be exposed to average out to a middle tone (I call it middle gray, but it could have a color tint). The metered area of this is a factor too (next page). Incident meters are a different story (second next page).
This is nothing new. Cameras meters have always worked this way. Photographers have always had to learn this. Here is an oldie but goodie Kodak Tech article: Accurate Exposure With Your Meter. Still very true and valid and fully applicable in every way. One quote from it says:
All of this has been well known for many decades, and reflective metering has always worked this way. It is good info, simply about how things work, and since it is obviously true, you can believe it. :) I promise, it will help you greatly if you know how to use this.
Sorry if it adds complication, but the attempt at right tries to show my meaning of "averaged" tone. There are minor technical inaccuracies now, but the overall concept is OK. It is the same rose picture, and imagine the round spot is the center area that is metered. Here, the color there has been "averaged" with the Photoshop Average blur filter, to show the numeric average value of those center pixels - dark ones and light ones and red ones, all blended together to be one average color. The camera meter uses a Gaussian blur filter, to be a bell curve that weights center more than the frame edges (more gradual fall-off than this sharp edge). This area has this middle tone because the pixels there were averaged, and then metered so exposure makes the average come out as a middle tone - not too dark, not too bright.
Something like this is all that reflective meters can do (an averaged tone is set to be about mid scale). And fortunately, a middle tone is often near about correct (most usual scenes do have wide range which does average near middle tone, which is how the system works). But the meter cannot distinguish those that do from those that don't. The meter can only see the light reflected from the scene, which is a lot of light if from white or light colors (so exposure will be turned down), or not much light if reflected from black or dark colors (so exposure is turned up). The camera cannot tell the difference, it just sees some light, so all it knows is to make all pictures average out a middle gray (the only possible goal - not too dark, and not too bright). So photograph something near all black, or something near all white, and both will come out near middle tone too, neither of which is correct in those cases. Simply how things work, and good to know.
The effect of a large area of light color is that the light colors become middle tones (darker, underexposure). The effect of a large area of dark color is that the dark colors become middle tones (lighter, overexposure). The only possible goal is middle tones. You can repeat this same type of test too, and you should repeat it, so you can see and believe it, and understand it, and be able to use it. This is indeed a LOT to know.
Newbies seem to imagine the camera ought to always be correct (imagining that our attention is not required). And maybe the meter is "correct" on all of them, but the meter may be doing something different than you expect. It only does what it can do. The meter has absolutely no clue what any of this scene is, no human recognition of what it means, and no comprehension how it ought to be. The meter cannot distinguish a rose from your Aunt Martha. Our human brain does have smarts and experience to know immediately, what it is, and how it ought to be, so that is our natural expectation. But the meter only sees a blob of light, which it can measure, but without any understanding about what it is, or what it means, or how it should be. So necessarily, what the result will be is a middle tone average, not too dark, not too light.
Those who imagine their camera meter is recognizing and evaluating details in our picture, and should always give the correct exposure, are simply in for big disappointments. Instead, the trick is in learning what the meter actually does, so we can use it as a good guide. The meter just gets it into the ballpark, often about correct, but from which we make relative adjustments, as we see needed. We humans can see the scene too, and in many cases, learn to recognize how it will come out, and can compensate in advance. That was a required skill with film, and is very handy with digital too, but digital shows us the result that we get, giving us another chance to fix it. Now is the time to realize that the one in charge of your camera's exposures is YOU.
I am NOT saying a handheld meter will be more accurate than the camera meter. It may be, if it is a incident meter, but if it is a reflective meter, it won't be any better, if even as good as, the camera reflective meter. What I am saying is, the above is how reflective meters work, affected by the subjects colors (and camera meters are reflective meters).
Said again, the exposure you get (underexposure or overexposure) will depend on what you aim the camera at. I am trying real hard to get your attention. If you want to resist, then imagine me shaking you by your shoulders shouting "wake up". :) We really ought to know how the light meter works (and this is how it works). This is all also true in sunlight too, but flash differences seem more pronounced (due to inverse square law falloff). The trick to know is that Flash Compensation is how we control TTL flash. Flash Compensation will greatly improve your TTL flash pictures. Simply watch, and do what you see you need to do.
Again, if any doubts, then it is absolutely necessary that you repeat this, do and see this kind of test in your own situation. It is real. It is the big overall view of how things actually work. All you need to repeat this is a sheet of black paper and a sheet of white paper, and your camera as a light meter. Then you can see, and believe, and will understand how it must influence your procedures.
We hear novices complain about exposure, thinking that their meter is misbehaving, when they simply don't understand how reflective light meters work. They may have a little more to learn, but it is easy when we know. The meter's goal in life is NOT to give "correct" exposure. The dumb meter has absolutely no clue how to do that. It has no clue even what the scene is, and certainly not how it ought to be, so there is never any concept of a "correct" reflective meter reading. The meter's goal in life is simply to expose all scenes to create a middle tone average, however much that takes, for whatever it is, regardless if we will agree that middle tone is "correct" or not. The five rose scenes above have five different contents, with five different average values. The reflective meter's goal in life is to make all pictures average out to be a middle tone (speaking of the central metered area). Therefore, the TTL automation exposes each picture of a different scene differently, to make that middle tone result be true (the average of the metered area). So dark scenes are exposed more, and light scenes are exposed less, to achieve the same middle tone every time.
This is simply how reflective meters work (and it is a good thing to understand and expect this). Note that the photographer's job is to see and realize if this current scene is darker or lighter colors than normal, i.e., if this exposure is going to need our help to come out darker or lighter as it should. This is easy today, the digital camera shows the result to us, then and there, while we can still correct it. When you know why, you also know how, so this is easy, no big deal at all. You will soon already "just know" before you take the picture, what to expect, and what to do about it... it quickly becomes second nature. It does involve looking and thinking.
Two more cases of additional hard and obvious proof. You can and should repeat these simple tests yourself.
These three cards (two are fun foamies) really are black and white and gray, here propped up on a couch background. This combined overall scene more nearly averages out to actually be middle gray (averaging a middle tone) - more what light meters expect to see, so its exposure is relatively correct. Reflective light meters simply try to make every scene average out to middle tone. The overall average tone is not too dark, not too light.
Black card with TTL flash (result is not black).
The metering used much more flash power (exposure) to make the black card be middle gray (tone). It makes couch and pink paper be white.
White card with TTL flash (result is not white).
The metering used much less flash power (exposure) to make the white card be middle gray (tone). Makes couch black, and pink paper dark.
Gray card with TTL flash
Approximately correct. The small paper was in fact pink, and the couch looks normal. Middle gray is made to be middle gray (tone), which is merely coincidence here. All images are always made to be middle tone average - that is all a reflective meter can do.
Get some black and white paper (from craft store), or any dark and light objects - items around the house - white walls, or bed sheets, or refrigerators. Maybe a black suit jacket and other dark stuff. Try this yourself, to see it, to believe it, and to understand and expect it.
This is simply what reflected light meters do. It is very good to know and expect this.
That was flash above, but now here are the same cards, repeated in a lawn chair outdoors in bright sun (no flash), same thing. It works the same way. This subject is not just about flash, it is about reflective light meters (like in cameras). This set is all automatically metered (without adjustment) by D300 camera in A mode (aperture preferred), f/8 ISO 320 (full frame shown, not cropped). The first scene, of the three cards, does actually average middle gray overall, so it comes out correct. All four come out middle gray, which is what reflected light meters do.
Three cards, f/8, 1/1250 second, ISO 320. It is Sunny 16.
(-2 stops 1/320 to 1/1250 second, +2 stops f/16 to f/8)
Black card, f/8 1/200 second (result is not black)
Result is 2 2/3 stops overexposed, from first one.
White card, f/8 1/5000 second (result is not white)
Result is two stops underexposed, from first one.
18% gray card, f/8 1/800 second
Result is 2/3 stop over first one.
Any and all scenes will be metered to come out middle tone (average value of the metered area, not too dark, not too light). This is simply what reflected light meters do. It is very good thing to know and expect this. The exaggerated plain card background scene is used here to simply make it trivially easy to see this. These cards are not typical scenes. Any real scene will have mixed brightness areas (sky and trees and shadows, etc), so that even unusually light or dark real subjects will rarely need as much as the two stops compensation these two "unreal" all white and all black subjects need (to actually be white or black). But it is routinely true that using reflective meters, light colored subjects (higher reflectivity) likely will need 1/2 stop or 1 stop more exposure to make them be light. Dark colored subjects (less reflective) likely will need 1/2 stop or 1 stop less exposure to make them be dark. The closer to middle gray the subject averages, the more accurate the reflected meter reading.
The point is, assuming the common reflective meter, in the same light, what your meter reads depends on what you are metering. The reflective light meter will try to give an exposure that will make everything average out to about middle gray. "Average" or "typical" scenes/subjects generally do have wide tonal ranges which do in fact average out to about middle gray, but which is not true in all cases. But this is the basis of the system, and you, as the photographer, can see the subject, and are supposed to realize this. If your subject is a more typical one, with typical wide range, and it actually does average out about middle gray, then great, easy as pie. If not an "average" subject, then you better pay attention, and stand ready to help with Flash Compensation or Exposure Compensation.
In contrast, handheld "incident" meters also exist, which stand at the subject's position, and aim back at the camera, and directly read the incident light falling on the subject, totally independent of the color of the subject (page after next). Again, incident meters read the light directly from the subjects position, so cannot be built into the camera. This may be less convenient to use (except is wonderful for studio flash), but are greatly more accurate and consistent.
Bottom line: When the exposure does not come out right, it cannot help to cuss the meter or bemoan your fate. This is simply how meters work, and have always worked. The way we learn to use them is to study the picture, and figure out what the meter was metering to cause that middle tone result... why it happened? (usually, dark or light colored subjects or backgrounds are expected to cause mid-tone results). Then we learn to recognize those situations, and then we easily just know ahead of time what to do to prevent it next time we see a similar situation . There is no other way but to learn.