<p>In this paper, we propose a network evolutionary game model to study the coevolution of mask-wearing behavior and epidemic spreading over complex networks. By applying mean-field and pairwise approximations, we transform the stochastic coevolutionary process into a system of ordinary differential equations and derive the strategy dynamics and state dynamics. We show that in a stable equilibrium of the strategy dynamics, infected individuals do not wear masks, and susceptible individuals wear masks only if both the contact probability between susceptible and infected individuals and the benefit-to-cost ratio of wearing masks are high. Numerical simulations on random and scale-free networks show that over half of susceptible individuals are prone to wearing masks for self-protection in the early stage of the epidemic outbreak. As a result, voluntarily wearing face masks can effectively slow the spread of the disease and flatten the infection peak.</p>