Previous research suggested that socially anxious individuals may form threatening interpretations of positive facial stimuli. However, some studies failed to evidence an interpretation bias toward positive faces. These studies may suffer from several methodological limitations that the present study attempted to overcome, and none of them assessed the interpretations bias, if exist, over time. The present study tried to figure out whether the socially anxious individuals are characterized by an interpretation bias towards positive facial stimuli, and investigated whether this interpretation bias tends to occur at an early stage in the interpretation process of positive stimuli, or it occurs at a later stage. The current study consisted of a 2 (stimulus duration type) ×2 (group type) mixed factors design. The within-subjects factor of duration type had two levels (long vs. brief durations), and the between-subjects factor of group type had two levels (high vs. low social anxiety groups). Before the experiment, participants first finished a package of questionnaires (SIAS, SPS, BFNE), then they were told that they would conduct an impromptu speaking task after the experiment, and their performance would be assessed. Participants were made to convinced that the experiment is about to test their social and presentation skills. After the experiment, no speaking task would be performed. Participants were debriefed and explained about the design. In the experiment, a positive face and a matched neutral face were presented on the screen at the same time each trial. Participants scoring high and low on Social Interaction Anxiety Scale were instructed to imagine these faces represented the audiences in their coming speaking task, and they were asked to identify the facial displays which they subjectively perceived to be more threatening. The matched positive and neutral facial displays were presented at both long (unlimited) and brief (60ms) stimulus durations. By doing so, interpretation biases to perceiving threat in positive facial stimuli were assessed over time under the public speaking situation. The numbers of the positive facial displays chose to be more threatening when compared with the neutral ones were recorded by the program. The dependent variables were analyzed with repeated measure ANOVA. Results showed that high social anxiety participants chose significantly more positive facial stimuli than low social anxiety participants at long stimulus durations, while no differences between groups could be found at brief stimulus durations. We come to a conclusion that, in contrast with the control group, high social anxiety participants are characterized by an interpretation bias towards positive facial stimuli, and are more likely to perceive them in a more threatening way in social interaction. It also suggested that only at long durations, participants with high vs. low social anxiety tended to interpret positive facial stimuli in different fashions. So based on these findings, we then propose that this interpretation bias for positive facial displays in socially anxious individuals may occur at a later stage in the interpretation process of positive stimuli.