Active learning methods put their efforts on selecting and labeling the most informative examples out of a large amount of unlabeled ones. It is performed in uncertain environments where the learner is required to make some decisions on the observed examples. However, existing algorithms do not have a good formulation to evaluate the example's uncertainty by considering the inconsistency between conditional features and decision labels, while this inconsistency has been taken into account by fuzzy rough sets. Therefore, a fuzzy rough sets based active learning algorithm with stream based settings is proposed in this work. The lower approximations in fuzzy rough sets are used to compute the memberships of the unlabeled example, and the uncertainty is then used for decision. Experimental comparisons with other existing approaches demonstrate the effectiveness of the proposed algorithm.