Minimal affective learning is a phenomenon wherein people can learn about the affective meaning of other people with brief behavioral descriptions. Prior research mainly focused on affective learning with own-race faces. Own-race bias is a robust phenomenon describing that people can recognize own-race faces more efficiently than other-race faces. In the current study, we investigated whether own-race bias would influence minimal affective learning. Chinese participants learned Chinese and Caucasian faces paired with behaviors of different valence. After learning, they were asked to evaluate the learned faces and novel faces. Their eye movements and pupil diameters were continuously monitored during the experiment. We analyzed the change in pupil dilation to assess how much cognitive effort was required for affective learning. The results showed that participants only learned positive information with faces. Learning performance for other-race faces was similar with own-race faces. In addition, change of pupil dilation was larger when learning other-race than own-race faces, suggesting a greater cognitive effort for affective learning with other-race faces. Taken together, the results demonstrated that affective learning for other-race faces was more difficult than own-race faces. This research provided more support for the notion that different cognitive strategies were employed by faces of different race.