A 30-year (1970-1999) reanalysis of temperature and salinity is conducted by assimilating temperature and salinity profiles into an ocean model of the Baltic Sea with ensemble optimal interpolation approach. Some configurations of the reanalysis are presented. For example, the samples are chosen from the same season as the analysis time to address the strong seasonal variability. The impact of different observation time windows on the analysis results is also discussed. A locally determined alpha is adopted for the long-time-scale simulation. To assess the accuracy of the reanalysis, a set of comparisons between the reanalysis results and the free run results was performed. The root mean square deviations (RMSDs) between the reanalysis results and not-yet-assimilated observations at all levels show that, compared to the free run, temperature and salinity have been improved significantly, that is, by 31.1 and 38.8%, respectively. The vertical structure of the reanalyzed fields is also adjusted. The reanalysis results show that the improvements in both temperature and salinity are smaller at greater water depths. Comparison with independent CTD data, the reanalysis significantly improved temperatures and salinities in all layers relative to the free run. For temperature and salinity during the period of ship voyages, the RMSDs are reduced by 32.9 and 25.5%, respectively. The temporal variations of the deep-water salinity caused by saltwater inflows are better captured by the reanalysis than by the free run. Moreover, the reanalysis improved the estimation of the depth of the halocline and thermocline, which are overestimated in the simulation without data assimilation.