The present study aims to develop an efficient dynamic statistical model to describe the daily behavior of boundary layer ozone in Macau. Four types of Kalman-filter-based models were proposed and applied to model the daily maximum of the 8 hr averaged ozone concentrations within a decade (2000–2009). First, the boundary layer ozone was modelled with the timevarying autoregressive model of order p, TVAR(p), which is a pure time series model hindcasting the ozone concentration by a weighted sum of the ozone histories of the previous p days. Then, it was modelled with the time-varying autoregressive model with linear exogenous input, TVAREX-Lin, which combines the TVAR model and the exogenous input of key meteorological variables in a linear fashion. Next, the nonlinear TVAREX model (TVAREX-NLin) which assumes the nonlinear influence of individual meteorological variable on ozone was adopted. Finally, a semiempirical TVAREX model (TVAREX-O3) was proposed to address the coastal nature of Macau and the interaction between the input variables. It was found that the proposed TVAREX-O3 model was the most efficient one among the model candidates in terms of the general modelling performance and the capability of modelling the episode situation.