Machine learning has been extensively studied in transient stability assessment of power system. To ensure the conservativeness of assessment results, a transient stability assessment method based on support vector machine and decision function is proposed. Firstly, training samples are constructed by Monte Carlo sampling on the basis of the on-line operation mode of power grid at a certain time. Secondly, the input features are extracted by combining the quantitative evaluation and statistical theory of transient security stability with the initial feature set of power flow before fault. Thirdly, the parameters of SVM are determined by grid search and the correlation between input features and transient stability assessment results is trained to obtain decision values. Finally, the threshold values are determined according to the decision value of support vector to ensure the conservativeness of assessment results. The effectiveness of the proposed method is verified by New England 10-machine 39-bus and actual system.