Previous research on emotional bias in face perception has shown inconsistent findings, proposing either angry or happy faces to be detected more efficiently. A recent study showed that the anger superiority effect (ASE), which showed in the high attentional demand condition, vanished in the low attentional demand condition. The authors thus proposed an attentional demands modulation hypothesis to interpret the inconsistent findings. The present study tested this hypothesis in a visual crowding task in which participants were instructed to determine whether the target face was happy or angry. Attentional demands were manipulated by changing the strength of crowding, including presenting stimuli in different configurations (Experiment 1), and setting different target-flanker separations and presenting stimuli in different eccentricities (Experiment 2). Experiment 1 showed an ASE when the stimulus configuration incurred a high attentional demand. Intriguingly, the ASE became weaker and then disappeared as the attentional demand became lower. Experiment 2 replicated this finding and showed that the ASE decreased as the target-flanker separation became larger. Together, these results suggest that the emergence and magnitude of ASE is modulated by attentional demands, which supports the attentional demands modulation hypothesis.