Important aspects of the seasonal variations of surface ozone are discussed. The underlying analysis is based on the long-term (1990?2004) ozone records of Co-operative Programme for Monitoring and Evaluation of the Long-range Transmission of Air Pollutants in Europe (EMEP) and the World Data Center of Greenhouse Gases which do have a strong Northern Hemisphere bias. Seasonal variations are pronounced at most of the 114 locations for any time of the day. Seasonal-diurnal variability classification using hierarchical agglomeration clustering reveals 5 distinct clusters: clean/rural, semi-polluted non-elevated, semi-polluted semi-elevated, elevated and polar/remote marine types. For the cluster "clean/rural" the seasonal maximum is observed in April, both for night and day. For those sites with a double maximum or a wide spring-summer maximum, the one in spring appears both for day and night, while the one in summer is more pronounced for daytime and hence can be attributed to photochemical processes. For the spring maximum photochemistry is a less plausible explanation as no dependence of the maximum timing is observed. More probably the spring maximum is caused by dynamical/transport processes. Using data from the 3-D atmospheric chemistry general circulation model ECHAM5/MESSy1 covering the period of 1998?2005 a comparison has been performed for the identified clusters. For the model data four distinct classes of variability are detected. The majority of cases are covered by the regimes with a spring seasonal maximum or with a broad spring-summer maximum (with prevailing summer). The regime with winter?early spring maximum is reproduced by the model for southern hemispheric locations. Background and semi-polluted sites appear in the model in the same cluster. The seasonality in this model cluster is characterized by a pronounced spring (May) maximum. For the model cluster that covers partly semi-elevated semi-polluted sites the role of the photochemical production/destruction seems to be overestimated. Taking into consideration the differences in the data sampling procedure the carried out comparison demonstrates the ability of the model to reproduce the main regimes of surface ozone variability quite well.