Multibody system (MBS) simulations see an increasing relevance in the field of automated modeling strategies due to several reasons. A redundant set of coordinates in combination with constraint coordination equations are required for such modeling strategies. Consequently, the numerical challenge of solving such models is much higher than it might be if a minimal set of coordinates representation is used. The current dissertation will explain the development of a model order reduction (MOR) technique, which allows to decrease the number of equations, and increase the numerical efficiency of such redundant multibody systems. A databased MOR approach has been chosen since the focus is on a general applicability to arbitrary multibody systems. It is based on the Proper Orthogonal Decomposition (POD), which is adapted to the Special needs of multibody systems by: (1) Using velocity data of the MBS (instead of the commonly used position data). (2) Separate handling of each coordinate type due to its physical meaning (instead of the commonly used mixed coordinate approach). (3) Adding a residual term to the applied projection which ensures the initial conditions to be met. Due to the use of a data-driven reduction approach, the resulting reduction subspace includes constraint information of the original MBS model. Therefore, the present Dissertation introduces a constraint reduction method, which determines and eliminates redundant constraint equations of the reduced order model. In contrast to known literature regarding MBS reduction, the herein derived coordinate and constraint reduced order model is therefore always solvable and illconditioned. The efficiency of the MOR approach is outlined by several practical numerical examples, evolving from automotive tasks. The results underline the efficiency of the novel approach, which ensures high result consistency, while at the same time, the dimension of the mathematical model is reduced up to 90+%. Finally, the approach is also applied to parameter identification tasks in the context of multibody systems. In addition to the previous mentioned reduction, the adjoint equations are reduced as well, by projecting them onto the same subspace as the original MBS model.