Test oracle automation plays an important role in test automation. Many programs don’t have an oracle at the beginning of the testing, and the tester should verify all the software behaviors to check whether they are correct. Such the work is too heavy and time-consuming. In this paper, we present a efficient system to construct test oracle of the web applications using active learning and support vector machines. The system extracts the features of execution traces, then builds a predictive model to classify the passed traces and failed traces with a small training set. Our approach is reducing the human oracle cost by active learning and sampling strategies, and get high accuracy of predicted labels.