The process of labeling data can be work intensive, not least in the area of sound event classification. Active learning can ease this intensive process and still provide an acceptable performance by a classifier. In this thesis, the performance of the classification of sound events in an active learning setting is evaluated in terms of robustness w.r.t. bias in the data. An active learning algorithm for evaluating sound event data is built and an experiment is performed. The result is that with the used feature extraction method, there is no difference in performance between when active learning is used and when active learning is not used. Also, there is no difference between the sample selection methods random, margin and entropy selection in this scenario.