In structural engineering, it is necessary to design structures with incomplete knowledge of the creep and shrinkage characteristics of the concrete to be used. Therefore, prediction based on concrete strength and composition is required. After summarizing the criteria for a sound prediction model, we discuss in detail the theoretical justification of model B3, including the thermodynamic restrictions, reasons for using power functions, consequences of microprestress relaxation and of activation energy, problems of characterizing aging by strength gain, consequences of diffusion of pore water for size and shape effects on shrinkage and drying creep and their asymptotics, and separation of cracking effects. Then, we focus on unbiased fitting of the existing worldwide database, which is characterized by limited range and complicated by variable data density. We present a statistical evaluation of models B3 and B4 and their statistical comparisons to other prediction models, and we describe the procedure that was used for calibration of the constitutive parameters by fitting a combined database of several thousand laboratory curves of limited time range and of about seventy histories of excessive multidecade deflections of large-span prestressed bridges. Finally, we briefly mention analytical methods for prediction of creep and shrinkage via homogenization.