In conjunction with increasing computational power, numerical models gained importance in virtually all engineering sciences during the last decades. Sophisticated modeling approaches are employed to simulate the physical behavior of the investigated structures for many practical problems. However, many of these models do not match with data measured in real-world structures to a certain degree. These inaccuracies are often caused by parameters needed for sophisticated modeling being imprecise, determined using engineering judgment or even completelyunknown. Structural systems are a particular case with many unknown parameters due to varying environmental conditions, approximated loads and imprecise knowledge of boundary conditions. This thesis aims to modify parameterized numerical models of structural systems automatically to increase conformity between simulation results and measured data. This process is known as model updating from the relevant literature. The deviation between numerical results and measured data is minimized using optimization methods via modifications of parameters of the numerical model. A further aim is to apply the same methodology to damage localization. If a model building a good representation of measurement data is found using a first model updating step, the methodology is applied again after a damage event to localize these damages. This technique is useful in structural systems, where assemblies may be inaccessible, and ongoing visual inspections are likely to be costly. Different metrics formulated in both frequency and time domain are investigated for their application in model updating within this text. A new scheme for the automated adjustment of numerical models is presented, enabling a comprehensive view of methods and techniques needed to perform iterative model updating. The aim is to minimize the outcome of these metrics, which results in nonlinear optimization problems providing several local minima. A new two-step algorithm employing state-of-the-art optimization methods is introduced to minimize the metrics, constituting a possible implementation of the scheme. The two steps consist of the global optimization method Simulated Quenching and the local Sequential Quadratic Programming. The need for a global algorithm is demonstrated using the concept of convexity and practical examples. Both methods can handle constraints to keep parameters within a specific user-defined range. Since the algorithm is a random procedure, it is started multiple times. A method to employ the objective function value to distinguish correct from wrong solutions is demonstrated. The distinctive metrics and their performance for updating different structures are investigated using a study on a simulated wind turbine in operation, a model of a three-story frame, a real scaled 34m wind turbine rotor blade and a scaled prestressed concrete tower. Although the examples focus quite heavily on assemblies from wind turbines, all concepts and methods introduced in this thesis are designed to be applied to general structural systems. A transfer of the methods to other physical processes is conceivable, although these applications may require different sensor settings. The classical approach to deviation quantification between model and measurement, using eigenfrequencies and mode shapes, performs good for most examples, but it has the drawbacks that the consideration of nonlinearities in the numerical model is not straightforward and it is not as sensitive to structural changes such as damage. Transient analyses are needed to account for nonlinearities, yielding time series that can be compared directly to measured data. Advantages and disadvantages of metrics for various structures are discussed in detail.