The use of a realistic prediction model for creep and shrinkage properties is an essential requirement for good design of structures sensitive to creep and shrinkage. With this aim, the lecture begins by discussing development of an extended, statistically unbiased, database on concrete creep and shrinkage, and of the statistical approaches to comparison and evaluation of various existing prediction prediction models. The proposed statistical method separates the scatter in the database due to variability in concrete composition and curing from the scatter in the evolution of creep and shrinkage in time. The necessity of using bias compensating weights and statistics based on the method of least squares and on the criterion of maximum likelihood is emphasized. Statistically correct mutual comparisons of the existing prediction models are then presented. They conflict with the comparisons and model rankings in ACI new guide 209.2R-08, which are non-standard, statistically questionable and misleading. The importance of calibrating the prediction model by short-time measurements of creep, shrinkage and water loss measurements for the given concrete is underscored. The second part of the lecture attempts a critical appraisal of the structural analysis methods. The long-term prediction errors caused by ignoring the cracking damage, the nonuniformity of drying, and the statistical uncertainty, are highlighted. The usefulness and the misleading potential of accurate solutions according to the linear aging viscoelasticity is then appraised in this context. The statistical analysis of the effects of creep and shrinkage in structures is discussed, the need to calculate the 95% confidence limits of structural effects is emphasized, and conclusions are drawn. In the conference presentation, examples showing comparisons with various observations on structures, including the ill-fated world-record prestressed box girder in Palau, are presented.