Rapid growth of parasites like Varroa destructor is one of the main reasons for elevated mortality of bee colonies. Beekeepers have to perform time consuming manual sampling to enable treatment and avoid colony losses. Most existing sampling plans only produce rough estimates and can be invasive and costly. This can be a significant stress factor, when considering an average sample size of 300 bees per apiary, to get a significant test result. This yields the question, if it would be possible to automatically monitor the infestation status of a beehive, using a non-invasive method. This works provides a first step towards answering this question. Therefore a camera system capable of creating continuous recordings of the entrance of an apiary is designed with whom more than 7TB of video data is recorded. From the conducted video material, a ground truth dataset is created with more than 13,000 manually labeled images of infected and healthy bees. The dataset is used to train and evaluate two detection approaches: A “traditional” machine learning pipeline and a deep learning pipeline using convolutional neural networks. The final evaluation shows that distinguishing between healthy and infected bees is possible using the convolutional neural network approach, providing the proof of concept. A per-class classification accuracy of 94.4% for healthy bees and 85.5% for infected bees is recorded with an overall f-score of 0.82, calculated on the labeled test dataset. This work therefore provides a novelty approach for automatic parasite classification and represents as a first step towards automated parasite monitoring.