Abstract: In the process of IC design and test, black-box modules, uninitialized sequential units, clock-domain interface and erroneous behavior of analog-to-digital converters, which may lead to the emergence of unknown values(X). The transmission of X-value in the circuit will seriously affect the activation and sensitization of faults and decrease the test coverage. In this paper, a method of predicting test coverage loss based on extremely randomized trees is proposed to the case where multiple inputs of the circuit are X-values. The data set is extracted by performing simulation analysis, circuit partition, and extracting structural features. Then train high-precision prediction models to quickly analyze the loss of circuit test coverage under multi-input with X-value. The experimental results show that the average prediction accuracy is 94.47% and 21.72% higher. The minimum prediction result of a single circuit is 89.03%, and the maximum value is 99.99%, which indicates that the prediction model has good stability.