The current experiment examined how set size of study array, face feature types (internal feature only vs. both internal and external features) and race played a role in a Delayed Match-to-Sample Probe Recognition Task. In this task, participants first studied a set of faces (two or four faces with either only internal features or both internal and external features) for 900 ms. And then a face with the same type of feature as in the study array appeared in the center of the screen. Next, participants had to press a key to judge whether or not the probe was a face that appeared in the study array (50% of trials). The probe remained on the screen until a response was detected. In this study, visual working memory capacity was estimated by calculating Cowan’s K for each participant in each condition based on the formula of Cowan (2001): K = L × (H - FA), where K is an estimate of the number of remembered faces in visual working memory, L is the number of faces presented in the study array (two or four faces), H is the hit rate and FA is the false alarm rate in the match-to-sample probe recognition task. Mean K-values were submitted to a repeated-measures ANOVA considering set size (two vs. four), face type (face with internal features vs. face with both internal and external features) and race (Chinese vs. Caucasian) as within-subjects factors. The results indicated visual working memory capacity estimated by Cowan’s K increased for four faces than two faces were displayed, and participants had larger working memory capacity for faces with both internal and external features than for faces only with internal features. These results suggest that the set size and face features may increase visual working memory capacity. However, there was no difference between own race faces and other race faces were displayed. People can retain the same amount of information in visual working memory for own- and other-race faces. Furthermore, results indicated interactions of race, set size, and feature type. For own-race (Chinese) faces, visual working memory capacity estimated by Cowan’s K increased for faces with both internal and external features in two faces condition, but no effect for four faces condition. For other-race (Caucasian) faces, visual working memory capacity increased for faces with both internal and external features in both two- and four-face conditions. Therefore, these results suggest that the processing of own- and other-race faces in visual working memory still differs although the visual working memory capacity is the same for own-and other-race faces.