Organisms are expected to be sensitive to cues of genetic relatedness when making decisions about social behaviour. Relatedness can be assessed in several ways, one of which is phenotype matching: the assessment of similarity between others' traits and either one's own traits or those of known relatives. One candidate cue of relatedness in humans is facial resemblance. Here, I report the effects of an experimental manipulation of facial resemblance in a two-person sequential trust game. Subjects were shown faces of ostensible playing partners manipulated to resemble either themselves or an unknown person. Resemblance to the subject's own face raised the incidence of trusting a partner, but had no effect on the incidence of selfish betrayals of the partner's trust. Control subjects playing with identical pictures failed to show such an effect. In a second experiment, resemblance of the playing partner to a familiar (famous) person had no effect on either trusting or betrayals of trust.