What is "ISO" in general, and how is the scale defined?
How does the ISO scale for film speed differ from ISO sensitivity as used in digital cameras?
Is lower ISO always better?
In short, the actual letters ISO are a name for the International Organization for Standardization (not, officially, an acronym -- more information here), and in photography it refers to the ISO 12232:2006 standard and other related standards: ISO 12232:1998, ISO 5800:1987, ISO 6:1993 and ISO 2240:2003. (Links on the Film Speed page.) Film historically has also used the ASA and DIN standards, the former using the same numbering system, and DIN using an entirely different scale.
The standards for film and digital are technically different (in ways I haven't investigated closely enough to report fully on), but generally they're similar enough that for practical purposes, they're essentially the same (notwithstanding Reciprocity Failure, which many films are quite prone to, though digital generally is not). So if you measure an exposure with your digital camera, you could use that exposure with a film of the same rating as the ISO setting you used in the digital camera, and expect to get a similar exposure (unless the shutter speed is long or short enough for Reciprocity Failure to kick in for the film) (also assuming similar equipment otherwise -- no differences in filters, etc.).
In both digital and film, a higher number indicates greater sensitivity. A number twice as high is twice as sensitive (e.g. 200 is twice as sensitive as 100, 400 twice as sensitive as 200, etc.). So, when shooting in relatively low light, and wanting a relatively-fast (e.g. fast enough to stop motion) shutter speed, a higher sensitivity rating will be essential (so no, lower is not always better!)
In digital cameras (and similarly but differently in film), higher ISO ratings tend to create noise (the related effect in film is increased graininess). So, while it's not always better (that depends on what you're going for), lower ISO ratings are always (or at least almost always) lower in noise, which may be desirable. (In the case of low light photography where shutter speed is not a concern, long exposures mixed with lower ISO ratings will create a "better" image -- though it's conceivable that some may like the effects of the noise; certainly there's appeal at times to film grain.)
As for how the scale is defined, it is based on measurements of an image produced based on a certain scenario of illumination. The details are complex, so I'll leave the details as an exercise for the reader. A lay summary is that (for digital) it's a measure of how quickly the digital sensor becomes "saturated" with light. (For film, the process is related but different.)
In summary: higher ISO is more sensitive but noisier (but not necessarily worse), digital and film rated for the same ISO (or ASA for the film) will have similar sensitivity, and the scale is based on how quickly an image will become "saturated" given a certain amount of illumination.
NOTE: I'm gearing up to do some experimentation related to the controversial answer from Matt Grum. Hopefully, my results will create a nice clear noise-free answer to the important point he brings up: that a high-ISO image with a low amount of light will be less noisy than a low-ISO image with the same amount of light getting to the sensor, which is later amplified in post-processing. More to come, hopefully in... EDIT: Well, I've failed to make this happen for a while now. I may still do it at some point. In the mean time, I'll also point to this article that talks about comparisons of native versus non-native ISO values, and amounts of noise in them, which, while the article doesn't exactly say so, I think is probably directly related to exactly this question.
TL;DR: To minimise noise, get as much light as you possibly can through the lens (since noise is mainly caused by low light levels) by using appropriate aperture and shutter speed settings.
Then while keeping these settings constant, increase ISO as far as you can (without clipping highlights) as this will reduce the proportion of the signal that is read noise.
Is lower ISO always better?
For a fixed amount of light coming into the camera, lowering the ISO will not result in a reduction of noise (improved signal-to-noise ratio). The only way to reduce noise is to combine lowering the ISO with letting in more light by opening the aperture or leaving the shutter open longer.
If the amount of light you can let in is limited (you have hit the max aperture and can't use a longer shutter speed without introducing blur) then using the highest ISO possible (without clipping highlights) will lead to the lowest noise in the image. This is going to seem counter-intuitive and invite downvotes (seven so far!), so please let me explain (see my response to comments at the end).
High ISO values don't cause noise, lack of light causes noise. The reason people associate high ISOs with noise is because when you increase the ISO setting in shutter or aperture priority mode, the camera either closes the aperture or increases the shutter speed to compensate, both of which decrease the amount of light coming into the camera.
Here's an explanation of why low light results in noise:
One important source of noise in images arises from the random nature of light and is called photon noise, or shot noise. Photons are emitted randomly from light sources. If you collect a lot of photons the randomness averages out and you get almost exactly the same number emitted in each direction. If you collect much fewer photons the number collected in neighboring pixels (that should see the same color) can differ giving brightness variations referred to as noise or grain. This is how lack of light leads to noise. See Wikipedia: Shot noise.
Another source of noise is read noise. Read noise happens when the analogue charges on the sensor are digitized (read out). Read noise is approximately constant with respect to the number of photons captured. Increasing ISO amplifies the signal, and thus the photon noise, but the read noise stays the same.
If you use a lower ISO (with the same amount of light coming into your camera) you will get an underexposed image and when you brighten it in post you will amplify both the photon noise and the read noise. Your total noise will then be higher.
Here is an example of this effect in practice:
These images were shot with identical settings bar ISO, and processed in exactly the same way. The bottom one is clearly noisier, despite being shot at ISO 100.
If you were to plot noise vs ISO for every pair of shutter-speed/aperture, you'd find noise always increases (sometimes only slightly) as ISO decreases. Therefore I don't see how you can argue that high ISO results in noisier images, given that when all else is equal lower ISOs give more noise!
Response to comments
Sorry, but your explanation doesn't change the fact that your basic statement is wrong. What you normally want to use is the lowest ISO that will give proper exposure with a usable combination of shutter speed and aperture
The method you use is of no real consequence, when you think about it using auto-mode and picking the ISO value that results in the shutter/aperture being as long/wide as you can get away with is equivalent to just setting the shutter/aperture to be as long/wide as possible in the first place and then setting the ISO to obtain the correct exposure.
However, I wasn't suggesting a shooting methodology, the point I was making was, to minimize noise, you want to let in as much light as possible, given that you have achieved this with shutter/aperture/flash or whatever, the ISO should be as high as possible (before highlights clip), as anything else will result in underexposure and thus more read noise.
if you actually do change your shutter speed and/or aperture (following the law of reciprocity, as one would normally do), then guess what: you get more photons on the sensor for the same "exposure" (call it EV as adjusted for the ISO in question)! So, less noise!
Yes but you can't increase the ISO and at the same time change settings to let in less light and then conclude you have more noise because you're using a higher ISO. Especially when you can show that if you increase the ISO on it's own you decrease noise!
It's like buying a car with a smaller engine in order to reduce fuel consumption, driving everywhere with your foot flat to the floor and concluding that small engines increase fuel consumption! Change one variable at a time.
Regarding the statement:
Is lower ISO always better?
There seem to be a variety of opinions on this topic, and while they may seem mutually exclusive, I am not certain that is the case. There is no cut and dry "Yes, X ISO setting is always better." I think which is better is very dependent on context...on what it is you are trying to shoot, and what kind of light is available.
The two points of view expressed so fare include:
The general consensus is that opting for the lowest ISO possible to achieve a proper exposure is the best approach. That statement is heavily laden with hidden meaning, however, as the lowest ISO possible may not necessarily be ISO 100. You may be forced to use a higher shutter speed or a narrower aperture to achieve the necessary creative effects you desire, forcing you to use a higher ISO to maintain proper exposure. You may also face issues with available light and reach the limits of your lens (i.e. maximum aperture) and be forced to use a higher ISO to achieve a correct exposure. I think that is the point Matt Grum has been trying to make.
Barring any specific creative needs (i.e. action-freezing shutter speed or large DOF via small aperture), the lowest ISO setting with the necessary shutter and aperture to produce a "correct" exposure (i.e. an exposure that is neither over or under exposed, or if you follow ETTR, one that has not blown out any highlights) is, still, the best practice. This can be easily demonstrated, as the sample shots below show. This set of examples is shot with several things in mind:
The clearest shot is, as you can see, the ISO 100 shot. At maximum aperture, ISO 100 required a 0.6 second exposure time. This is pretty long, but since there was no motion, a lengthy exposure is not an issue. The ISO 1600 shot is still properly exposed, and used a shutter speed exactly four stops faster than the ISO 100 shot. Despite being properly exposed, there is obviously additional noise. The final shot was another ISO 100 shot at the same settings as the ISO 1600 shot, with the exposure value adjusted by +4EV in Lightroom. The noise level in this shot is obviously FAR higher than the other two since it has been digitally enhanced.
Given these examples, we can come to a few logical conclusions:
I think a good way to learn what ISO settings are best for various scenes and lighting is to set your ISO to Auto, use manual mode, take a few shots, and review the results. The camera's automatic exposure metering will always try to create a "proper" exposure, and when you have a scene with a broad range of tones, it should choose the right setting most of the time. You can also try manually setting a higher or lower ISO than the one the camera chooses automatically, and re-take a shot to see the results. Outside of maybe landscape and still life photography, you probably won't find a single "correct" or "best" ISO setting. However, I do believe the general rule of thumb will always be:
Use the lowest ISO you can while maintaining a proper exposure for the type of shot you are making.
For still life, that will probably always be the minimum native ISO (not using any kind of ISO expansion). For landscape, that will probably be lower ISO's, such as 100 or 200. For action photography, including sports, wildlife, birds, kids, etc., the lowest ISO may change for every shot, and may vary from ISO 200 through ISO 3200 or beyond, and it will be very dependent upon available light. Lots of light will allow you to use lower ISO, less light will dictate higher ISO. Regardless of what ISO you use for action shots, another good rule of thumb is:
Its always better to actually get a shot than miss one because you don't like the camera settings necessary to capture it.
Even if you have to use ISO 3200 to get decent indoor sports shots with your f/1.4 lens, at least you'll get the shots. Those ISO 3200 shots will have lower noise than the ISO 1600 shots that you underexpose then correct via post processing, as shown by the (rather extreme) example above. De-noising algorithms are also pretty advanced these days, and can greatly reduce the noise level of a high ISO shot to more acceptable levels. That again makes using a higher ISO that will ensure a correct exposure a better option than a lower ISO that will likely underexpose and require post-process correction.
Further investigations to cover Matt's assertions about ISO have lead to another sample image. The image below is split into four bands, two representing exposure following Matt's methodology, and two representing exposure where ISO is kept as low as possible. The assertion was that when trying to expose a scene, you set aperture and shutter then use the highest ISO possible without blowing out highlights, and correct exposure via post processing by reducing EV, to produce an image with the lowest noise. This contradicts the normal assertion that one should use the lowest ISO and adjust aperture and/or shutter to achieve a correct exposure to maintain the lowest noise possible.
The first and third bands in the image above were shot by choosing a specific shutter speed and aperture, then upping the ISO as far as possible without clipping highlights. The second and fourth bands in the image above were shot by choosing ISO 100 and a specific aperture, and adjusting shutter speed to achieve a correct exposure (no ETTR.) Both images were corrected in post production by using Lightroom's "Auto Tone" feature, which caused a reduction in the high ISO image's exposure by a little bit, and pretty much left the ISO 100 image the same.
There is a difference of three stops ISO between these two shots, and the increased noise level of the ISO 800 image is very clear in the shades. In the midtones, there is some observed increase in noise in the ISO 800 image over the ISO 100 image. In all tonal levels, fine detail has been lost to one degree or another in the ISO 800 image vs. the ISO 100 image. This can be observed by examining the remote control in its cradle, the fronts of the palm, and in any shadow tones. In the mid tones and highlights, noise levels are not specifically high enough to cause any significant degradation in print. The noise level in midtones and shadows is, however, likely to intrude onto fine details, such as one might pick up with macro photography or any photos that push the limits of resolution of a lens or sensor.
While negative exposure compensation in post processing has indeed reduced the noise level of the high ISO image to more acceptable levels, there is no question that using the lowest ISO possible will result in lower noise than even post-process negative exposure compensation of a high ISO image.
The real question is, is it always appropriate to use the lowest ISO possible? If you can expose the scene you are trying to capture without any undesirable side effects, such as blurring, underexposure, etc. then choose the lowest real ISO you can (expanded ISO settings usually achieve a lower ISO setting through digital, rather than analog, means...so using ISO 50 Expansion for example, should be avoided.) If you can not expose the scene you are trying to capture without any undesirable side effects, which might be the case when photographing sports or wildlife, photographing concerts, or doing much of any kind of indoor photography involving any action, then increasing the ISO to the minimum acceptable level that will allow you to expose your scene properly (i.e. eliminate motion blur, expose at the right level, etc.) should be chosen.
If you have the headroom, overexposing by choosing the maximum ISO you can without clipping highlights, and applying some negative exposure compensation in post processing, can help mitigate the effects of noise of very high ISO, and will be a better choice than choosing an ISO that is too low, then applying some positive exposure compensation in post processing (which will just heighten the effects of noise.)
ISO is effectively a sensitivity of the sensor, whether it be film or digital. In theory, ISO for a digital camera should be the same as for a film camera.
The ISO on film is determined by the grain size of the chemical. What this will mean is that the resolution will be better with a lower ISO film. Also, because a film grain is all or nothing, this will lead to some spottiness, or noise.
The way that ISO is achieved in a digital camera can help to understand why it is important, and why using a higher ISO isn't always a good thing. I don't know the way it was achieved in film, so I can't really speak to it, but I imagine it's using somewhat similar principals.
First of all, Matt did an excellent job explaining the different sources of noise. What happens in a camera actually comes down to the following.
The camera then may remove some of the basic sources of noise, especially the "Dark current" noise. That's the signal if you were to expose complete blackness.
So what actually happens when you turn up the ISO? There's basically two paths, which I'll cover below.
If the ISO is increased, and the shutter speed correspondingly decreased, then the signal level going into the A2D converter remains the same. However, the amount of signal on the sensor is decreased. That means that any noise which is sensor depended will in effect be amplified. This includes things such as Shot Noise. Some other forms of noise won't be amplified.
The second option is if the original image was underexposed, but the ISO brings it to being correctly exposed. It is possible to contrast enhance the underexposed image to have the same max/min as the correctly exposed image. The signal coming out of the A2D converter will effectively have a smaller range. This will leave only a few light levels, leading to a much noisier looking image than the correctly exposed image.
A few other interesting things. Sensors tend to bleed if they are saturated to neighboring pixels. If an image is increased in ISO and decreased in shutter speed, then it will not be saturated on the FPA itself, removing some artifacts. This can lead to somewhat better performance when saturated.
In general, high ISO on a digital camera will have less noise then the same ISO on film, due to a much better way to achieve high ISO.
Hopefully this wasn't too technical, but I'm just excited to share my knowledge of Electrical Engineering with the community:-) Let me know if you have any more questions, and I'll try and explain it even better.
To answer the second question, I'd say that lower ISO isn't always better, but in general, it's best to use the lowest ISO you can. If you are shooting indoors, for example, a higher ISO will allow you to use a shorter exposure time, leading to less motion blur. Usually I use the lowest ISO that will allow me to not be over the hand-limit of 1/lens_length shutter speed. But there are always exceptions.
In analog photography, ISO measures the sensitivity of the film to the light. A film with higher speed will reach its saturation (or overexposure) point faster.
You can use a low-ISO film in bright light. In dimmer situations, it may require a wider aperture than you have available or a very long shutter speed.
You can use a high-ISO film when there is not enough light for proper exposure. However, if you use it with lots of light, you will have to restrict the light coming to your camera with shutter speed and aperture.
(Not well-versed in analog photography so not much info here.)
In digital photography, ISO is basically this:
It is just an arbitrary1 number that describes the amount of amplification applied to the analog2 signal from the sensor before converting it into a digital number. It represents how many photons are required to produce a value of "one" on the resulting digital pixel.
It does not describe the sensitivity of the sensor.
Increasing the ISO will not let you see more light in dark conditions and it will not saturate the sensor any faster. It will just stretch the signal so less light is required to produce the same pixel value, also amplifying the noise. So the image will saturate faster, but you will not be using the full potential of the sensor.
So if ISO is amplifying the noise, is less always better? No. If you do not have enough light, the only alternative to high ISO is amplifying the image in post-processing, which will amplify all the same noise you'd get with ISO. Even worse, it will add even more noise.3
Then where does the common misconception that more ISO equals more noise come from? Because people, being used to analog, would pick an ISO number and stick with it, and then set the exposure to prevent overexposing the resulting image.
This is upside-down because you are restricting the amount of usable light entering the sensor to satisfy an arbitrarily chosen constraint.
Instead, pick the best aperture and shutter speed for your conditions. This will minimize shot noise. Then raise ISO as much as possible without overexposing, to reduce the amount of read noise.
Or just let the camera choose ISO for you automatically.
1 They do set the numbers to make the image look like an image taken with a film of that ISO speed under similar conditions. Unfortunately, they don't take into account sensor size which adds even more confusion to the common myth that larger sensors are better in low-light conditions.
2 Some cameras cheat by also amplifying the digital signal, after A/D conversion. This doesn't help with the myth that higher ISO produces more noise, when the opposite is true in reality.
3 The additional noise is called read noise, which is introduced by the A/D converter after applying ISO.
In digital photography, ISO speed rating does not characterize sensor sensitivity. ISO speed setting on a digital camera is controlling the amount of amplification / multiplication of the signal from the sensor after the data is already captured. ISO speed does not control the sensitivity of the sensor, and is more like a "push" processing of a film, thus controlling overall brightness of the outcome.
Because of the limitations of the analogue-to-digital converters in some cameras, higher ISO may result in less noise, and less plugged shadows.
What I have missed in all the answers so far is that ISO is basically equivalent to film's ASA rating by describing a material constant describing the relation of luminous exposure (the time integral of per-area luminous flux density) to saturation of the medium (film negative or sensor/file maximum values on digital). If you work with larger sensors/films/plates while retaining the same ISO/ASA value, you will need physically larger apertures in order to cover the larger area with the same brightness. For the same framing, you'll have proportionally larger focal length, so it turns out that ratios like f/2.8 combine well with ISO values and exposure time in order to figure out correct exposure.
Now a film has grain, and film with higher sensitivity will achieve its results at the cost of a higher grain size. In a similar vein, higher sensitivity for digital sensors comes at the cost of larger pixel sites, resulting in either reduced pixel counts or requiring larger sensors. The highly regular grid of a digital sensor and comparatively high efficiency compared to film actually makes another factor relevant: noise. This comes as amplification/quantification electronic noise as well as statistical photon registration noise (the latter is to some degree relevant with film since it also is subject to the likelihood of photon/grain interaction). You can store film in darkness for months without change but you cannot run an electronic sensor in the dark without getting a signal.
Now for a significant amount of digital sensors, there is variable (ISO-depending) analog amplification of the sensor signal before digitization so the ISO choice has a permanent impact on your raw image's content. If I remember correctly, Sony's "Exmor" branded sensors don't have this kind of analog gain so their raw image data is independent from the ISO setting and ISO is more a matter of the postprocessing applied that scales that data to some desired brightness.
So basically, the higher the pixel counts, the more noise you can expect for higher ISO values, and larger sensor areas will help in countering that. Distributing more complex electronics for CMOS sensors over larger areas also helps with keeping heat production for continuously used sensors (required for mirrorless cameras and/or live view or video) low, and sensor heat is another source for noise.