The occurrence of haze events is the consequence of complex interactions between aerosols and specific multi-scale processes, including atmospheric boundary layer variations, weather and climate processes, and atmospheric chemical processes. Utilizing the hourly observation data of PM_2.5, surface visibility and relative humidity in Chengdu, China, from 2014 to 2017, the atmospheric extinction coefficient time series of six haze events during the observation period were retrieved firstly by Koschmieder's formula. Then based on the phase space reconstruction theory, the optimal delay time τ and the optimal embedding dimension m of the corresponding series were calculated, and the recurrence plot method was used to verify that the atmospheric extinction coefficient time series of the haze events originated from the evolution of the deterministic systems. Finally, the time series nonlinearity of atmospheric extinction coefficient in haze events was tested. The analysis of nonlinear characteristic variables (saturation correlation dimension, maximum Lyapunov exponent and Kolmogorov entropy) showed that the time series of atmospheric extinction coefficient had weak chaotic characteristics, the application of Cao method excluded the possibility that the time series was random, and the further diagnosis of surrogate data method finally determined that the extinction coefficient time series of the haze events were nonlinear. Our findings reveal the complexity of the atmosphere extinction coefficient time series of the haze events and lay the foundation for a deeper understanding of the evolutionary behavior of haze systems.