Existing adversarial face detectors are mostly developed against specific types of attacks, and limited by their generalizability especially in adversarial settings. In this paper, we propose a new detection strategy based on a dual-classifier driven deep-feature disentanglement model for detecting different types of adversarial faces. Experimental results over adversarial examples and face forgery attacks show that the proposed detection method is effective with better generalizability and more adversarially robust comparing with previous methods.