​An experience of an afterimage is caused by a previously seen stimulus, when that stimulus itself is no longer present.

Negative afterimages exhibit inverted lightness levels, or colours complementary to, those of the stimulus and are usually brought on by prolonged viewing of a stimulus. They are best seen against a brightly light background. They occur (as least in part) because some cells (cones) on the retina do not respond to the present stimulation because they have been desensitised by looking at a previous stimulus.

(By contrast, positive afterimages are the same colour as the previously seen stimulus. They often occur when there is no stimulation—for example because the lights have gone out, or because your eyes are closed and your hands are in front of them to block all light. In these conditions they occur when some cells (the cones) on your retina keep transmitting signals to the brain for a little while after they have been stimulated. But they can also happen in other conditions, such as when presented with a previously seen outline of a shape, as occurs in the Colour Dove Illusion.)
Wheel showing complementary colours opposite each other. e.g. blue is complementary to yellow, and red to cyan.

Negative afterimages can be relatively complex as the inducing flag image below attests. Stare at the bottom right corner of the yellow rectangle for 30 seconds to one minute and then look at a white surface, such as a blank screen, a white wall, or a piece of paper. Blinking a couple of times while looking at the white surface may help you to experience the afterimage.
Figure to induce an afterimage of an American flag

And some afterimaging inducing stimuli can produce surprisingly photo-realistic afterimages, such as that below.
An understanding of the physiological mechanisms behind negative afterimages requires a brief discussion of the photoreceptive rods and cones which reside in the retina and are the light sensors of the visual system. Rods and cones are two distinct types of specialized neurons (information-carrying cells of the nervous system) responsible for phototransduction, the process of converting light energy in the form of a photon into electrical energy. The electrical current is then transmitted via the retinal ganglion cells (RGCs) whose axons (protruding nerve fibers which carry the outgoing signals) form the optic nerve which relays the signal to the brain.
A single photon (light particle) can activate the light-sensitive photopigment molecule in a rod, resulting in a signal from the eye to the brain. This extreme sensitivity of rod photoreceptors makes them best suited to low lighting levels and so they are responsible for our night vision (scotopic vision). Rods are too sensitive to mediate fine discrimination between wavelengths, and so provide only ‘black and white’ vision, and this is why it is difficult to distinguish colours on a dark night.
Cones are far less sensitive than rods – while a single photon may activate a rod, a cone must absorb around 100 photons to produce an equivalent response (Purves et al. 2010, Ch.10). At around the level of twilight, both rods and cones are functional (mesopic vision); however, cones dominate in daylight (photopic conditions) whereas rods are saturated and contribute little to vision. Cones also have the advantage of being far less convergent than rods; ‘midget’ retinal ganglion cells (RGCs) near the centre of the retina receive input from a single cone, whereas larger ‘parasol’ RGCs at the periphery may receive signals from over 1000 rods (Frishmann 2005).  This means that cones enjoy greater spatial resolution or acuity - the small region of the retina which produces the sharpest image (the fovea) is populated almost exclusively by cones. Most importantly, each cone belongs to one of three varieties pertaining to the particular kind of light-sensitive photopigment molecule it contains. Each pigment has its own distinct absorption spectrum and so cones mediate our colour vision. White daylight is the sum of multiple wavelengths (colours) and that most black and white images result from cone activity—on the other hand, a very dim flash of coloured light at the threshold of vision for a dark-adapted subject will appear colourless because only the rods will function. So, most negative afterimages that we encounter result from cone signals.

Rods and cones each contain many millions of their specialized photopigment molecules. When these molecules are struck by light they undergo a process called bleaching (a change in both structure and colour) which generates a flow of current in the form of sodium and calcum ions, before being eventually restored to their original state. Negative afterimages are often said to result from adaptation in the rods and cones (Schiller & Tehovnik 2015); when the retina is exposed to light for any period of time, the sensitivity of the activated photoreceptors in that particular region will decrease. The mechanisms underlying this adaptation are not entirely understood, but it appears to involve a form of negative feedback from the amount of photopigment bleaching that has taken place. This feedback regulates the level of signal gain when a photon is absorbed by a photoreceptor (Baylor 1996; Purves et al. 2010). The cause of negative afterimages now seems to be as follows: if part of the retina is subjected to pure green light, the M-cones in that area will receive more stimulation than the S- or L-cones. Faced with a subsequent uniform white stimulus (i.e. mixed wavelengths), that part of the retina which has been most desensitized to green wavelengths will produce a stronger output from the S- and L-cones relative to the rest of the retina —hence one will experience negative afterimages, appearing in approximately complementary colours. Thus a red stimulus will produce a green afterimage (and vice versa), a blue stimulus will produce a yellow afterimage (and vice versa) and a black stimulus will produce a white afterimage (and vice versa), and so on.

This explanation is at least part of the explanation of the occurrance of negative afterimages, however, recent evidence suggests that adaptation of the cortex also has a role in the production of negative afterimages, in addition to the cells in the eye. Shimojo et al. (2001) found that fixating a neon-color spreading illusion led not only to negative afterimages corresponding to the lines forming the image, but also to one corresponding to the perceptually filled-in surface during adaptation. This negative afterimage of the filled-in surface was not caused by a corresponding stimulus affecting the eye.

Negative afterimages appear to move with one's eyes as they are caused by the effects described above on the retina. Occasionally one might be fooled into thinking that a negative afterimage is a real patch of colour in the world. This can happen when one looks at a plain white wall. Sometimes one and has an experience as of a dirty mark on the wall when there is none there. In fact, one might be experiencing a negative afterimage. One can usually tell whether it is an afterimage by seeing whether the apparent mark moves with one's eyes or stays at the same location of the wall.

Many philosophers of perception seek to analyse afterimages as pathological cases of visual experience. Many philosophers think that in order for an experience as of seeing an object to amount to genuine visual perception, that object must exist and one’s visual experience must be appropriately caused by that object. Now, visual illusions are usually analysed as cases in which one perceives the objects of the public, external world, but that perceptual experience is somehow inaccurate or non-veridical. In the case of an afterimage, one may have a lasting visual experience as of a green square on a white wall, but no such green square, or indeed a square of any sort, need exists independently of one’s own nervous system in front of one at that time (and, in the case of negative afterimages, no green square need to have been previosuly in front of one). Afterimages are not ordinary public objects, but rather they arise as artefacts of individual perceptual systems. This has led many philosophers to suggest that the visual experience of an afterimage is a failure of perception, and hence afterimages are best characterised as a type of hallucination.

However, this line of thought can be resisted. One might try to argue that one is seeing the square inducing figure—albeit seeing its colour incorrectly, and one is also seeing it at a later time that it was present in front of one. Such delayed perception might seem odd, but, in its defence, one could point to the fact that there are other cases of delayed perception. For example, when we see the stars in the night sky, we are looking at the stars as they were years ago. Indeed, a star that we may happily say that we presently see may in fact no longer exist. It may have blown up in a supernova, the light from which has not yet reached us.

Philosophers disagree as to how we should best explain illusions and hallucinations, and some theories of perception may accommodate one phenomenon better than they do the other. See Macpherson (2013) for a detailed overview of various philosophical approaches to hallucinatory perceptual experiences. Afterimages figure in debates as to whether we are directly aware of physical objects or, rather, internal (mental, private) objects called sense-data. Those who think that afterimges are hallucinations and who hold that, when we have a visual experience as of an object, we must be aware of an object, typically hold that the experience of an afterimage involves the experience of a mental object (a sense-datum), as there is no such physical object to be seen (see Robinson 1994; also see Crane and French 2015 for discussion).

Many philosophers, particularly physicalists, reject this conclusion and argue that experiences of afterimages can be given other explanations. Jack Smart’s famous paper in which he argues for physicalism contains a discussion of the experience of an afterimage (Smart 1959). Smart claims that when we experience an afterimage we only seem to be aware of an object, but we are not. He says we can explain away those appearances as follows:

'When a person says ‘I see a yellowish-orange after-image’ he is saying something like this: "There is something going on which is like what is going on when I have my eyes open, am awake, and there is an orange illuminated in good light in front of me"' (p. 150)

Many philosophers, however, have claimed that this analysis is an insufficient account of what is going on in the experience of afterimages. See, for example, David Chalmers (1996, p. 360).