Arecalibration of the B3 model for creep and shrinkage has become possible with the availability of a new expanded statistical database on creep and shrinkage at Northwestern University that now also covers high performance concretes with admixtures. The recently collected evidence on excessive multi-decade deflections of numerous pre-stressed long-span box-girder bridges shows that this recalibration is indeed necessary. Data containing precise information of concrete materials as well as curing and testing conditions are needed in order to update the model parameters.To accurately capture the long-termbehavior of a structure, the model must also be calibrated with long-term data.Yet, precise laboratory tests pertain mostly to durations <6 years and are not available on the time scale of many decades that engineers need to predict the behavior with desired lifetimes >50 years and often>100 years.To performthe required recalibration, two databases are used simultaneously- one for laboratory test data with accurate material information but unknown long-term behavior and another for long-term bridge deflections with unknown or uncertain material information and structural characteristics. A Bayesian approach to this problem is presented. First the parameters of the B3 model are recalibrated with the enlarged laboratory test database, which is regarded as the Bayesian prior information, see Figure 1. Then the model is updated in the Bayesian sense using the incomplete bridge deflection data to obtain the posterior information. The resulting model predicts the long-term structural behavior more accurately and provides the associated measures of uncertainty based on both sets of data.