A keystone species is one whose effect on its ecosystem is disproportionately large relative to its abundance. The concept reorganised how ecologists think about which species matter, and it has been criticised for being applied far more loosely than the evidence supports.
Robert Paine coined the term in 1969 after work on the rocky shore of Washington state. He removed the predatory starfish Pisaster from experimental plots and left adjacent plots alone. In the plots without the starfish, mussels outcompeted everything else and species richness collapsed from around fifteen species to roughly one. The starfish was not abundant, and removing it transformed the entire community.

The borrowed image is architectural: the keystone at the apex of an arch bears little of the weight, and removing it brings the arch down.
Sea otters in the North Pacific eat sea urchins. Urchins eat kelp. Where otters were hunted out for the fur trade, urchin populations exploded and grazed kelp forests down to bare rock, known as urchin barrens, destroying habitat for a great many other species. Where otters returned, kelp forests returned with them. This is both a keystone example and the standard illustration of a trophic cascade.

Keystones are not only predators. Ecosystem engineers such as beavers physically create habitat; fig trees fruit when little else does and sustain frugivores through lean seasons; and prairie dogs and other burrowers shape whole grassland communities.
The term is applied too freely. Paine intended a specific claim demonstrable by removal experiment. It is now applied to charismatic species on the basis of plausibility rather than evidence, and critics argue that a concept covering everything explains nothing.
Context dependence. A species that is keystone in one place, or one season, may not be in another. Pisaster itself does not have the same effect across its whole range.
It may reflect study effort. Species that are easy to observe and easy to remove experimentally are more likely to be identified as keystone, which is a bias in the evidence rather than a fact about ecosystems.
Conservation consequences. Designating a species as keystone directs funding and attention, which can be efficient or can starve less charismatic species that matter as much. Some ecologists prefer to speak of functional roles rather than of particular species holding systems together.

The underlying finding is solid and important: ecological effect is not proportional to abundance, and losing an uncommon species can restructure a community. The dispute is about how confidently the label can be applied to any particular species without the removal experiment that justified it in the first place.