Mimicry--when one organism (the mimic) evolves a phenotypic resemblance to another (the model) due to selective benefits--is widely used to illustrate natural selection's power to generate adaptations. However, many putative mimics resemble their models imprecisely, and such imperfect mimicry represents a specific challenge to mimicry theory and a general one to evolutionary theory. Here, we discuss 11 nonmutually exclusive hypotheses for imperfect mimicry. We group these hypotheses according to whether imperfect mimicry reflects: an artifact of human perception, which is not shared by any naturally occurring predators and therefore is not truly an instance of imperfect mimicry; genetic, developmental or time-lag constraints, which (temporarily) prevent a response to selection for perfect mimicry; relaxed selection, where imperfect mimicry is as adaptive as perfect mimicry; or tradeoffs, where imperfect mimicry is (locally) more adaptive than perfect mimicry. We find that the relaxed selection hypothesis has garnered the most support. However, because only a few study systems have thus far been comprehensively evaluated, the relative contributions of the various hypotheses toward explaining the evolution of imperfect mimicry remain unclear. Ultimately, clarifying why imperfect mimicry exists should provide critical insights into the limits of natural selection in producing complex adaptations.