Histogram
A histogram is a graph counting how many pixels in the image fall at each brightness level, from pure black on the left to pure white on the right.
Also known as: exposure histogram, tonal histogram
The histogram is a graph that counts how many pixels in the image fall at each brightness level, from pure black at the left end to pure white at the right. Unlike the screen, whose apparent brightness changes with the light where you happen to be, the histogram measures the file, not perception.
The horizontal position is the tone and the height is the number of pixels that have it.
A frequent mistake is looking for a centered bell curve. There is no correct histogram: a night shot's piles up on the left, a beach's on the right, and both can be perfectly exposed.
The horizontal scale isn't linear in light either: the axis is divided into output tones, and in a RAW file the right half of the graph holds most of the available levels. That is why a photo exposed one stop under loses more information than it seems when you lift it in processing.
How do I read it in three seconds?
Look at the edges first and the mass afterward. The edges tell you whether you have lost information —that is clipping and it can't be recovered—; the mass tells you whether the exposure matches the scene in front of you.
Histogram shape: Mass on the left not touching the edge; Scene that produces it: Night, low key; Is there a problem?: No; What to do: Nothing
Histogram shape: Spike stuck to the left edge; Scene that produces it: Underexposure or empty shadow; Is there a problem?: Yes; What to do: +1 EV, or raise ISO
Histogram shape: Mass on the right not touching the edge; Scene that produces it: Snow, beach, high key; Is there a problem?: No; What to do: Nothing
Histogram shape: Vertical spike against the right edge; Scene that produces it: Blown sky or reflections; Is there a problem?: Yes; What to do: −1 EV, or graduated filter
Histogram shape: Two mountains with a valley between them; Scene that produces it: Interior with a window, backlight; Is there a problem?: It depends; What to do: Bracket and blend
Histogram shape: Everything concentrated in the center; Scene that produces it: Very overcast day, fog; Is there a problem?: No; What to do: Contrast in processing
There is one exception that isn't a flaw: light sources. A streetlamp, the sun or a specular reflection will be clipped in any reasonable exposure, and lowering the whole photo by two stops to save them ruins the rest.
Should I look at the luminosity or the RGB histogram?
The RGB, whenever you have the option. The luminosity histogram averages the three channels and hides the most common case of clipping: a saturated red or blue that hits the top while the overall brightness looks right. On a red flower in the sun, the red channel clips long before luminosity does.
Another warning: the histogram you see on the camera is calculated from the JPEG preview, not from the RAW. With a neutral or flat picture profile, that preview looks more like the real file and the histogram lies less. For how to take advantage of the margin left on the right, see the entry on ETTR.
Common mistakes
Exposing to center the bell curve. Applied to a night scene, that means two or three stops of overexposure and a gray sky.
Looking at it only after shooting. In an electronic viewfinder and in live view the histogram is visible before the shot, which is when you can still correct.
Ignoring the highlight warning. The blinking areas show where the clipping is, something the histogram doesn't locate in the frame.
Mistaking a tall spike for a problem. A uniform sky generates a huge spike in one specific zone, and that only means there are many pixels at that tone.
Judging color by the luminosity histogram. A white balance cast shows up in the separation between the three channels, not in the combined curve.
What to look at in your next photo
The two edges, in this order: right first, because clipped highlights don't come back, and left afterward. If neither is stuck to the edge, the exposure is usually recoverable. To read it while you shoot, see the guide to reading the histogram, the article how to read the histogram while you shoot and the interpretive use in creative histogram.