Evidence for the effect of eye images on prosocial behavior is mixed: some studies have found that images of eyes enhance cooperative behavior while others have not replicated this effect. In a recent meta-analysis, Nettle et al. (2013) argued that previous null results have occurred because data have been analysed incorrectly. Specifically, the authors predicted that eye images either (i) reduce variance in donation amount by reducing the likelihood that the Dictator will keep the entire endowment or (ii) increase compliance with cooperative norms. Since several previous studies have only looked at the effect of eye images on mean donation amount, they have not tested these predictions. We test both hypotheses here using a Dictator Game (n = 779) conducted over Amazon Mechanical Turk, a setup that has previously yielded negative results. We provided players with two types of descriptive norm information, by telling them how much others typically give in this setting. We compared donations under the norm treatments with decisions made under control conditions, where no norm information was supplied. In each setting, subjects were additionally shown images of eyes or a control image (flowers) so that any additional effect of eye images on prosocial behavior could be examined. Eye images did not increase Dictator Game giving, regardless of whether 'giving' was defined as mean donation or simply the tendency to make a non-zero donation. Furthermore, eye images did not increase compliance with a descriptive norm in this setting. Due to these negative results, we conclude that the conditions and underlying mechanisms that lead to positive effects of eyes on prosocial behaviour remain elusive.