The histogram is the digital descendant of the densitometer and the exposure-placement habit. Where you once read densities off a step wedge and decided where a tone should sit on the characteristic curve, the rear screen now hands you the whole tonal distribution at a glance. The instrument is faster, but the judgement is the same one Ansel Adams described in The Negative: the meter and the graph tell you where tones fall, and you decide where they belong. Read this way, the histogram earns its place in a film photographer’s thinking rather than replacing it.

What the axes really are

A histogram is a bar chart of tonal distribution. The horizontal axis runs from black on the left to white on the right; the vertical axis counts how many pixels carry each tonal value. In an 8-bit rendering that axis spans 256 levels, from 0 to 255.

The horizontal axis is not linear with respect to scene luminance. The displayed values are sRGB-encoded, the encoding defined in IEC 61966-2-1. That transfer function is piecewise: below a linear threshold (encoded value at or under 0.04045, corresponding to a linear signal at or under 0.0031308) the curve is a straight line of slope 12.92; above it the relationship is a power law, ((R' + 0.055) / 1.055) ^ 2.4, using the constants 1.055 and 0.055 and a decoding exponent of 2.4. Overall the curve sits quite close to a pure gamma of 2.2, with the short linear toe near black holding down quantisation noise where the eye is most sensitive to it.

The figure people quote, 0.45, is the encoding exponent (the OETF), the reciprocal of the roughly 2.2 decode gamma, and the two are routinely conflated. The mechanism that matters: gamma encoding redistributes a linear signal so that perceptually even steps through the midtones occupy more code values than the highlights do. That is why midtones spread across the centre of the graph while the brightest stops crowd toward the right edge of the display, even though, in the raw signal, the truth is the exact opposite.

Why the data crowds the right edge

The sensor’s response is linear: double the light, double the recorded value. Tonal levels are therefore distributed by stop in a way that has nothing to do with the gentle roll of the display. The brightest stop occupies fully half of the available levels, the next stop down a quarter, the one below that an eighth, and so on. A 14-bit raw file holds 16,384 discrete levels, so the top stop alone accounts for around 8,192 of them, the next around 4,096, halving down until the darkest stops share a tiny handful. A 12-bit file has only 4,096 levels to begin with, and the same halving applies.

This is the real argument for exposing to the right, introduced by Michael Reichmann on the Luminous Landscape in 2003 after discussion with Thomas Knoll, who wrote Adobe’s raw conversion. Push the exposure as far right as the highlights allow and the shadows are recorded with more photons. The benefit that actually counts, as Emil Martinec showed in Noise, Dynamic Range and Bit Depth in Digital SLRs (2008), is signal-to-noise ratio, not the number of quantisation levels: read and photon noise already dither the signal across several levels, so the theoretical advantage of “more levels in the shadows” is largely moot. ETTR buys you cleaner shadows because it captures more light, not because it fills more bins.

Why the screen histogram misleads the raw shooter

The histogram on the screen is not computed from the raw sensor data. It is derived from the embedded JPEG preview the camera builds from the current picture settings, and that preview has already been tone-mapped, contrast- and saturation-adjusted, and white-balanced. The mechanism behind the mismatch is the white-balance multipliers: to correct colour, the camera scales the raw channels by different amounts, typically pushing the red and blue channels above 1 while leaving green near 1. The tone curve and saturation then lift luminance further. All of this drives the JPEG values up toward clipping while the underlying raw channels still have room.

The magnitude is not small. In a documented high-contrast case on a Hasselblad X2D, Jim Kasson found that you must underexpose by 1 2/3 stops from the true ETTR exposure before the in-camera histogram stops warning of blown highlights (How to Expose Raw Files – Part 2, Lensrentals, May 2023). In the same scene, the blue and green raw channels were still about a stop and a half from clipping when the JPEG histogram already showed highlights pinned against the right wall. Expose to satisfy that histogram and you have thrown away the better part of two stops of shadow signal-to-noise for no reason.

A workflow that reads the raw file

The fix is to make the displayed histogram track the raw data. UniWB, unity white balance, forces the white-balance multipliers to roughly 1 so the embedded-JPEG histogram follows the raw channels instead of running ahead of them. The cost is cosmetic: because the raw green channel is no longer scaled down, the rear screen takes on a strong green cast. You learn to ignore the colour and trust the position. Off the camera, dedicated tools read the raw histogram directly: RawDigger for analysis and FastRawViewer for culling both show the true raw distribution rather than the JPEG’s tone-mapped version, so you can confirm exactly where each channel sat.

Clipping, shadows, and where you place a tone

Two failures still read straight off the ends of the graph. When highlight values pile against the right wall the pixels have reached full-well and recorded nothing: they are clipped, and no recovery returns what was never captured. A spike hard against the left wall is blocked shadows, crushed to black. Between those walls a modern sensor holds something like 13 to 15 EV of dynamic range, far more than the 8-bit 0–255 display can show at once, which is why raw shadow lifting can pull detail out of regions that look empty on the screen, up to the point where read noise swamps the signal.

A distribution leaning left or right is not in itself an error. Meter a snow scene and the reflected-light meter, calibrated to render whatever it reads as middle grey, will place the snow at Zone V, the meter’s calibration point. That point is middle grey at about 18% reflectance in classic Zone System terms, though reflected meters are in practice calibrated nearer 12 to 12.7% under ANSI/ISO, the origin of the long-running 18-versus-12.7 argument. To keep the snow white you place it deliberately at Zone VII, two stops up, which puts its peak roughly two-thirds of the way to the right edge of the histogram, not against it.

Here the parallel with film becomes a real difference, not just an analogy. The digital sensor has a hard right wall; past full-well there is nothing. The black-and-white negative does not. The characteristic, or H&D, curve of a black-and-white film such as Ilford HP5 Plus rolls off gradually rather than ending in a wall, so highlights compress smoothly and retain separation well above the metered point instead of slamming into a ceiling (Adams, The Negative; Lambrecht and Woodhouse, Way Beyond Monochrome). That shoulder is why overexposing a negative is forgiving and overexposing a raw file past the right edge is fatal. The histogram tells you where the tones have fallen. You, as on the curve, decide where they belong.