“Grain” and “noise” are used interchangeably to describe the fine texture that breaks up a smooth tone, but the two arise from unrelated physical processes. Silver-halide grain is a permanent structure built into a developed negative; sensor noise is a statistical fluctuation in the count of photons and electrons at the moment of capture. The distinction predicts how each behaves across the tonal scale, how it scales with enlargement, and how it reads on a monochrome print.

From Latent Image to Clump

A black-and-white emulsion is a suspension of light-sensitive silver-halide crystals in gelatin. Exposure does almost nothing visible: it builds a latent image. The accepted account is the Gurney-Mott mechanism, set out by R. W. Gurney and N. F. Mott in 1938. An absorbed photon frees an electron inside the crystal; that electron reduces a mobile interstitial silver ion to a neutral silver atom; and a cluster of roughly four such atoms — the Ag4 speck — is the minimum stable site that a developer can act on.

Development then performs an enormous amplification. The developer reduces the entire crystal that carries a developable speck to a tangle of metallic-silver filaments, turning a handful of atoms into a grain of the order of a billion atoms. That gain — on the order of 10⁹ — is the physical origin of film’s effective speed: a few captured photons commit a whole crystal to becoming visible silver. The visible texture is not one crystal but a clump, where neighbouring developed grains overlap and the eye integrates them into the irregular pattern seen at magnification. Crucially, this structure is fixed once the negative is fixed and washed; it does not change with how the print is later exposed.

Reading the Granularity Numbers

Manufacturers quantify the result as diffuse RMS granularity: the root-mean-square fluctuation in optical density, measured with a microdensitometer through a 48-micrometre circular aperture on film developed to a net diffuse density of 1.0, read at 12x magnification, in the reference developer D-76 (Ilford’s equivalent is ID-11) at 20 °C. The published figure is that RMS density fluctuation multiplied by 1000, so a value of 16 means an actual RMS density deviation of 0.016. Lower is finer.

Kodak’s own datasheet F-4017 lists conventional cubic-grain Tri-X 400 at 17 (fine) (the sheet-film Tri-X 320 comes in slightly finer at 16). The tabular-grain films are markedly finer: T-Max 400 sits at 10 and T-Max 100 at 8 under the same conditions (F-4016). The intuition behind the number is aperture averaging. Drag a small fixed window across the developed image: where silver gathers in a few large clumps, the density swings sharply from window to window, giving a large fluctuation and a high number; where grains are small and evenly spread, each window catches a similar amount and the fluctuation is small.

The tabular, or T-grain, emulsions Kodak introduced in 1986 owe their fineness to crystal shape. The crystals are flat plates with a high diameter-to-thickness (aspect) ratio; lying flat in the coating, they scatter less light and present more surface area per unit silver, so the emulsion is finer-grained at a given speed. Ilford’s Delta Professional films reach for the same end by a different route — patented Core-Shell crystals, a silver-iodide core wrapped in silver-bromide shells, rather than Kodak-style plates. Ilford does not publish RMS granularity for the Delta line, characterising grain through MTF and sharpness data instead, so any Delta-to-T-Max comparison is qualitative, not a number-for-number match.

Why Grain Climbs the Tonal Scale

Film grain is most conspicuous in the mid-tones and upper values, and the clear base of the deepest shadows carries little visible structure. The mechanism is Selwyn’s law (E. W. H. Selwyn): RMS granularity grows in proportion to the square root of mean density. Where there is no developed silver there is no fluctuation, and granularity rises as density builds up the tonal scale. Selwyn also showed that the granularity multiplied by the square root of the scanning-aperture area — G × √A, the Selwyn granularity S — stays essentially constant across aperture sizes (verified over a range from roughly 7.5 to 384 micrometres). That constancy is why a single number measured at one aperture predicts the texture at others, and why grain is a property of density rather than of where you happen to put the window.

Why Sensor Noise Does the Reverse

A digital sensor counts photons, and photon arrival is a Poisson process: variance equals the mean, so the shot noise of a count N is √N and the signal-to-noise ratio is N/√N = √N. A mid-tone photosite that collects 10,000 electrons therefore carries shot noise of √10,000 = 100 electrons, an SNR of 100. A shadow photosite holding 100 electrons carries shot noise of 10 — an SNR of only 10. Quadrupling the exposure (four times the photons) doubles the SNR. This is a property of light, present in a theoretically perfect detector.

A second component, read noise, is added by the electronics that amplify and digitise the charge. It is independent of exposure — typically about 10 to 20 electrons per pixel at room temperature, a few electrons in cooled scientific CCDs — so it sets a fixed floor and dominates only in deep shadows, where the photon signal has dropped below it. The span between the saturation point and that floor is the dynamic range, commonly expressed as 20·log₁₀(full-well capacity ÷ read noise); full-well capacities run on the order of 20,000 to 600,000 electrons for typical photosites. So the texture climbs the opposite way to film: cleanest near full well in the highlights, noisiest in the shadows.

Geometry differs too. Most sensors carry a Bayer colour-filter array, two green photosites for each red and blue (2G:1R:1B). Reconstructing a full-colour value at every pixel — demosaicing — interpolates between neighbours, which correlates the otherwise-independent noise spatially. The result can read as a fine grid or as chroma mottle, a regularity quite unlike the organic placement of silver clumps, and it persists into the file even after conversion to monochrome.

What You Can Control

Take Tri-X 400 in D-76 diluted 1+1 at 20 °C for about 9¾ minutes, a 35mm frame of 24×36 mm enlarged to a 12×16-inch print — roughly 12x linear. The emulsion grain is fixed at development; the enlarger simply magnifies it about twelvefold onto the paper, with nothing left to negotiate. Your one real lever is the developer. A solvent fine-grain developer such as Ilford Perceptol or Kodak Microdol-X dissolves grain edges and lowers granularity, at a modest speed loss used undiluted; a high-acutance, low-solvent developer such as Rodinal (Adox R09) does the reverse, leaving sharp, well-defined grain that reads as crisp texture at that 12x. Choose the developer and you have largely chosen the grain.

Digital inverts the timeline. The shot noise is set at capture by how many photons you collected, so the equivalent lever is exposure — exposing to the right pushes the signal toward full well and maximises SNR before the read floor matters. But unlike grain, the result is not yet final: the noise can be denoised, smoothed or sharpened after capture, redistributed across the file at the cost of some detail. That is the heart of the difference. Grain is baked in at the moment of development and merely magnified thereafter; noise is partly negotiable, fixed in its statistics at capture but still editable on the way to the print.