Heart diseases are amongst the most common causes of death in the industrialized world. Since the cardiological system is very complex and hard to capture in its entirety, researchers are looking for indicators of its health. One of these is the heart rate variability (HRV), i.e. the variation of the time interval between two heart beats. It reflects many physiological processes which influence the rhythm of the heart. Since these influences of the generation of heart beats are non-linear, researchers use a visualization tool, the Poincaré plot, which has its origins in chaos theory, to analyze HRV. This method, also called Lorenz plot, became popular in the last 20 years. It gives a simple visualization of the heart's beat-to-beat behavior and can be used for various applications, e.g., to predict the mortality of patients with myocardial infarction. Numerous data models exist in order to automatically quantify Poincaré plots. The main objective of this master thesis is to implement 14 of the most common models in MATLAB and improve them where possible. Afterwards these models are tested with respect to their ability to differentiate between pathological and non-pathological heart beat recordings, compare them to statistical HRV-measures and examine the models' dependences on each other. The data are filtered via clustering algorithms and used in four different test cases. The first case is to test for data length sensitivity. Therefore, each model is applied to non-pathological and pathological data sets, with a stepwise reduction of their data length. The second case is an application of the models on pathological and non-pathological data sets, at a fixed data length, in order to do a deeper examination of them and their ability to differentiate between these data sets. For the third test, the models are used on data sets of subjects before and after arrhythmia treatment. The final test case is a comparison of younger and older healthy subjects via the data models. Although not all implemented data models showed significant differences between the tested data sets, some passed all tests, including one, which was improved for this thesis. With the calculated correlations, the number of models which should be considered for further research can also be reduced.