The own-group recognition bias (OGB) might be explained by the usage of different face processing strategies for own and other-group faces. Although featural processing appears in general to impair face recognition ability when compared to configural processing (itself perhaps a function of acquired expertise), recent research has suggested that the OGB can be reduced by directing featural processing to group-discriminating features. The present study assessed a perceptual training task intended to replicate Hills and Lewis' (2006) findings: we trained White participants to focus more on discriminating parts of Black faces, in particular the bottom halves of the faces, expecting a reduction of the OGB as a consequence. Thirty participants completed the training task, and visual patterns of attention were recorded with an eye-tracking device. Results showed that even though participants modified their visual exploration according to task instructions, spending significantly more time on the lower halves of faces after training, the OGB unexpectedly increased rather than decreased. The difference seems to be a function of an increased false alarm rate, with participants reducing response criterion for other-group - but not own-group - faces after training.