Deformable template models are a very popular and powerful tool within the field of image processing and computer vision. This thesis treats this type of models extensively with special focus on handling their common difficulties, i.e. model parameter selection, initialization and optimization. A proper handling of the common difficulties is essential for making the models operational by a non-expert user, which is a requirement for intensifying and commercializing the use of deformable template models. The thesis is organized as a collection of the most important articles, which has been published during the Ph.D. project. To put these articles into the general context of deformable template models and to pass on an overview of the deformable template model literature, the thesis starts with a compact survey of the deformable template model literature with special focus on representation, model parameter estimation, initialization, optimization and performance measures. The original articles - aligned a bit in notation and corrected from discovered spelling errors and other typos - are enclosed in the appendices. Compared to the literature one contribution is a general scheme for estimation of the model parameters, which applies a combination of a maximum likelihood and minimum distance criterion. Another contribution is a very fast search based initialization algorithm using a filter interpretation of the likelihood model. These two methods can be applied to most deformable template models making a non-expert user able to use the model. A comparative study of a number of optimization algorithms is also reported. In addition a general polygon-based model, an ellipse model and a textile model are proposed and a number of applications have been solved. Finally the Grenander model and the Active Appearance Model have been explored and some extensions are presented.