Little is known about what design elements in digital learning games enhance learning; especially in the case of child audiences. This study examines the effects of a learning game's visual design on perceived attractiveness and learning outcomes. We developed two visual designs for the game: one with supposedly high esthetic value and another with a low esthetic value. Participants (children between 9 and 11 years of age, N = 53) were randomly divided into two groups. Each group interacted with their assigned version for about 20 minutes and then evaluated its visual attractiveness without knowing about the other version. Then, they evaluated the attractiveness of the complementary version. As the next step, they evaluated both versions side-by-side. During the free-choice period, children could continue playing one of the game versions or a different game. They clearly preferred the high esthetic version in evaluations (d > 0.86) and in the free-choice period (62% preferred the high esthetic version of the target game), but this did not improve their learning outcomes (comprehension: d = -0.59; transfer: d = -0.16). Possible explanations of this effect are discussed in terms of cognitive load theory and cognitive-affective theory of learning from media.