High levels of ozone occur frequently in Taipei, with the peak concentration appearing at noon. Favorable meteorological conditions are extremely important in explaining the observed episodes at stations located in central Taipei. Liu et al. (Atmospheric Environment 24A, 1461-1472, 1990) have proposed that the existence of a simplified local land-sea breeze circulation helps to form a temporal stagnant environment during morning hours which is conducive to the photochemical production and accumulation of oxidants. In this paper, the authors focus on analysing the relationship between the morning stagnant condition and the local diurnal circulation, and hope to set up a proper procedure for episode forecast. Through a detailed comparison of the surface meteorological data observed on the ozone episode days and those collected on the low ozone days, it is found that intensified temperature increase, high percentage of low cloud cover, no precipitation, clockwise change of wind direction and a high percentage of low wind speed from sunrise to noon, are the crucial parameters accompanying the noontime high levels of ozone. These findings are further clarified through an intensive observation program during 1989. The analyses of the intensive observation data suggest that the morning stagnant condition occurs due to a transition of flow divergence after sunrise to a flow convergence at noon in the basin. The major causes of such a transition are the diabatic solar heating and an efficient city heat accumulation which is characterized by an enhanced temperature gradient between the surrounding mountains and the basin during hours of large sea-land temperature gradient, in addition to the nighttime radiative cooling. Significant convective activities are observed in the afternoon. All these findings are combined into a proposed procedure for ozone episode forecast, in which the identification of a favorable synoptic environment allowing a significant development of local circulation is the first crucial step toward a successful prediction.