A review of self-injurious behavior (SIB) indicates several biological variables that should be considered in future research. A current hypothesis, that SIB is reinforced by attention, receives support but inadequately accounts for evidence that SIB can arise in the absence of reinforcement, is often difficult to modify, and has features unlike those of conditioned behavior. Alternative possibilities are suggested. Neurological damage is common in populations displaying SIB and may produce inadequate pain perception or avoidance. Stereotyped behavior, also common in these populations, may also be related to SIB. Stress and isolation, both strongly correlated with SIB, may evoke this behavior by altering levels of biogenic amines and affecting the probability of aggressive responses. Finally, SIB may have been favored in evolution by group and kin selection, as the loss of a damaged individual may actually favor his genotype through promotion of relatives. These alternatives are not necessarily mutually incompatible nor do they preclude a degree of flexibility in SIB. SIB may result from the interaction of several variables that independently increase its probability.