Mutualistic interactions, such as those between plants and pollinators, are predicted to be prone to shifts to abandonment of mutualism, to cheating, or to extinction. Cheating of pollinators by plants has provoked controversy among evolutionary biologists since Darwin, who argued that it was impossible to evolve. Here I review the patterns and stability of cheating of pollinators by plants, focusing on the Orchidaceae. Review of current phylogenetic and character state evidence in selected clades of the Orchidaceae supported the hypothesis that cheating is ancestral to those clades, while mutualism has evolved repeatedly from cheating. No evidence is found for instability of cheating clades compared to mutualistic clades over evolutionary time, nor that pollinator generalization is associated with mutualism. I argue that to understand the stability of cheating within pollination mutualisms, cheating should be considered as a frequency-dependent process, suggest analogies between the evolution of cheating and exaggerated sexual traits in male animals, and suggest critical experiments that will elucidate how and why mutualism evolves.