In this paper, we propose to enhance the original Random Forest algorithm by introducing probabilistic nodes. Platt Scaling is used to interpret the decision of each node as a probability and was initially developed for calibrating Support Vector Machines. Nowadays it is used to calibrate the output probabilities of decision trees, boosted trees or Random Forest classifiers. In comparison to these approaches, we integrate the Platt Scaling calibration method into the decision process of every node within the ensemble of decision trees. Regarding the original Random Forest, the nodes serve as a guide to predict the path through the tree until reaching a leaf node. In this paper, we interpret the decision as a probability and incorporate more information into the decision process. The proposed approach is evaluated using two well-known machine learning datasets as well as object recognition datasets.