This paper shows that reliability can be considered as a rational measure of structural performance in the design of bridges and assessment of exiting bridges. The reliability analysis procedure is demonstrated on a representative sample of composite steel girder bridges. Load and resistance parameters are identified, limit state function for the whole bridge was formulated, and reliability analysis is performed for girders and structural systems. It was found that the system reliability is considerably higher than the girder reliability, in particular in case of uncorrelated girder resistances. The difference between girder reliability and system reliability can be considered as a measure of bridge redundancy. The calculated reliability indices for the whole bridge, psystem, are compared with the reliability indices determined for individual girders, pgirder. It was observed that the ratio of Psystem / Pgirder decreases with increasing pguder. It ranges from 1.6 for pgirder = 4.5 to 3.1 for pgirder = 2.0. It was also found that the correlation between girder resistances plays an important role and it can decrease psystem by 10 to 30% depending on the girder reliability, span length, and girder spacing. The paper shows that much can be gained by considering a bridge structure as a system.