Every digital camera has a setting called ISO, and almost every explanation of it is subtly wrong. Not wrong in a way that ruins your photographs, but wrong in a way that makes the camera harder to understand than it needs to be, and that leads people to reach for the wrong control at the wrong moment.
The confusion has a specific cause. The word arrived in digital photography carrying a meaning it earned somewhere else, and it no longer means that thing. Once you see where the number came from, what it does now becomes obvious.
For most of photography’s history, film speed was a measurement, and the thing being measured was a material.
The groundwork was laid in 1890 by Ferdinand Hurter and Vero Charles Driffield, two industrial chemists working in Widnes. They established that the relationship between light and the resulting silver density on a plate could be plotted as a curve, measured repeatably, and compared between manufacturers. That discipline became known as sensitometry, and it turned “this plate seems quicker than that one” into a number you could argue about with evidence.
Competing systems followed. Scheiner in Germany in the 1890s. Then DIN in 1934, using a logarithmic scale where each increment of three degrees represented a doubling. Then ASA in the United States in 1943, using an arithmetic scale where the number itself doubled. In 1974 the two surviving systems were harmonised into a single international standard, which is why old film boxes carry a marking like ISO 100/21°: the arithmetic figure inherited from ASA, the logarithmic one from DIN.
What matters here is not the alphabet soup. It is what the number referred to.
An emulsion’s speed was determined by exposing it to controlled amounts of light and developing it under specified conditions. For black and white negative film, the speed point was defined at the exposure required to produce a density of 0.1 above the film’s base level. That is a physical event happening in a physical substance. Silver halide crystals of a particular size and distribution, suspended in gelatin, responding to photons in a way that could be measured with an instrument.
So ISO 400 was a fact about the roll in your hand. Load it, and the camera’s job was simply to be told what you had loaded so that its meter could do the correct arithmetic. The dial on the camera did not change the film. It described the film.
Even then the standards were not purely a matter of physics. In 1960 the ASA revision reduced the safety margin that had been built into the calculation, and film speeds roughly doubled overnight without a single change to the chemistry. A useful reminder that the number always sat on top of a convention. But the convention described something real, and the something real was in the canister.
There is one sensor in your camera. It was manufactured with a fixed ability to convert arriving photons into electrical charge, a property called quantum efficiency, and no menu setting alters it. The silicon does not become more sensitive because you turned a dial.
To see what the dial does instead, follow the signal. Light lands on a photodiode, which accumulates charge in proportion to the number of photons it catches. That charge is converted to a voltage. The voltage passes through an amplifier. The amplified voltage is then handed to an analogue to digital converter, which turns it into the numbers that become your raw file.
The ISO setting sets the gain of that amplifier.
That is the whole mechanism. You are not collecting more light. You are multiplying a measurement of the light you already collected, before it gets digitised. On most cameras, values beyond a certain point are not even analogue gain any more, just digital multiplication, which is arithmetic you could perform yourself in software afterwards.
Two consequences follow, and they pull in opposite directions.
The first is the one everybody knows. Amplifying the signal amplifies the noise that came with it. The dominant noise in a normally exposed frame is photon shot noise, an unavoidable statistical property of light itself: photons arrive at random intervals, so any finite count of them carries uncertainty. Gain multiplies signal and shot noise together, in lockstep, which is why a high ISO image looks grainy no matter how good the sensor is. If you collected few photons, no amount of amplification invents more.
The second consequence is less well known and more interesting. Some noise is added downstream of the amplifier, by the electronics that read and digitise the signal. Applying gain early means the real signal arrives at that stage already loud, so the read noise represents a smaller proportion of it. This is why raising ISO in camera can produce cleaner shadows than shooting at base ISO and brightening the file afterwards. You amplified before the noise was added rather than after.
How much this matters varies enormously between cameras. Sensors with very low read noise are close to what is called ISO invariant, meaning it makes little practical difference whether you amplify in camera or in software. Others reward in-camera gain noticeably. Many recent designs use dual conversion gain, with two separate amplifier circuits and a switch between them at a particular value, which is why some cameras look conspicuously cleaner at ISO 3200 than at ISO 2000. Noise does not always rise smoothly as the number goes up.
One more thing gain cannot do. The photodiode has a fixed capacity, so the brightest signal it can record is fixed too. Mapping a smaller collected charge to the top of the output scale means everything above that point clips. Each stop of ISO therefore costs you roughly a stop of highlight headroom.
This is where the familiar exposure triangle breaks down. Exposure, in the strict photometric sense, is the amount of light reaching the sensor, and it is determined by scene brightness, aperture and shutter speed. Nothing else. ISO is not an exposure control. It is a decision about how to interpret the exposure you made. The triangle is a useful teaching device that gets one third of itself wrong.
It is also worth knowing that ISO 12232, the standard governing digital speed ratings, permits several different methods of arriving at the figure. Manufacturers have latitude. ISO 400 on two different bodies is not a promise of identical behaviour, only of approximately comparable metering.
If ISO is not an exposure control, the practical hierarchy changes.
Treat light collection as the actual job. Your goal at capture is to get as many photons onto the sensor as the scene and your intentions allow, without clipping the highlights you care about. Aperture and shutter speed are the only tools that do this. Every decision about noise is really a decision made here, several steps before the ISO dial becomes relevant.
Then set ISO to suit the file you want out. Once light collection is settled, gain determines how bright the resulting file is and how it is distributed across the available range. That is a genuine choice, not an afterthought, but it is downstream of the important one.
Find out what your specific camera does. This takes twenty minutes and is worth more than any general advice, including this article. Photograph a static scene containing deep shadow at base ISO, deliberately underexposed by one, two, three, four and five stops. Photograph it again at correspondingly raised ISO values with the same aperture and shutter. Brighten the base ISO files in software to match, then compare the shadows at full magnification. If they look similar, your camera is close to invariant and you can leave ISO low and lift later with confidence. If the raised ISO versions are visibly cleaner, in-camera gain is buying you something real. While you are there, step through the range and look for the dual gain point.
Use Auto ISO deliberately, not apologetically. Of the three settings, ISO is the one with the least creative consequence. Aperture governs depth of field, shutter governs how motion renders, and both are decisions about what the photograph is. Gain is mostly a decision about brightness. So it is the correct parameter to automate: set your aperture, set a minimum acceptable shutter speed, cap the maximum ISO at whatever you have decided you can live with, and let the camera solve the remaining variable. This is not laziness. It is delegating the least important choice.
Be sceptical of the extended values. The L and H settings beyond the native range are typically digital multiplication rather than analogue gain, sometimes with a highlight penalty at the low end. Anything the camera achieves by multiplying numbers, you can do yourself afterwards with more control and the option of changing your mind.
Stop thinking of high ISO as a solution to darkness. It is a way of making an underexposed frame usable, which is a different and more modest thing. When a shot is too dark, the honest options are more light, a wider aperture, or a longer exposure. ISO is what you reach for once those are exhausted and you would rather have a noisy photograph than none. That is often exactly the right trade, and there is nothing shameful about making it. It is just worth knowing that you are making it.
The through line is this. For eighty years the number on the dial described a substance you had bought and loaded, and your job was to report it accurately to the meter. Now the same number describes something you are choosing to do to a signal. The word survived the transition. The meaning did not.
