The loss distribution approach is widely used in operational risk estimation. While the detailed form of frequency distribution does not significantly affect risk analysis, the choice of model in terms of severity often has a significant impact on operational risk measures. Usually, different heavy-tailed distributions, which have their own characteristics, are respectively used to quantify operational risk. In order to integrate characteristics of different heavy-tailed distributions and to reduce the uncertainty of the operational risk model, we propose a combination model to estimate operational risk in this paper. The model has three stages. First, we estimate operational risk by using different heavy-tailed distributions in the loss distribution approach framework. Second, the weights are decided according to certain criteria, and in this paper, p-values of the Kolmogorov-Smirnov goodness-of-fit test are used to decide the weight of each operational risk. Finally, the results obtained in the previous phases are combined in an integrated estimation to form the final result. We use a linear combination model to measure the operational risk capital of Chinese banks. We also compare the results measured using the combination model with those obtained from the basic indicator approach. Empirical analysis shows that this approach allows capital to be allocated in a more efficient way than the standard approach (only one heavy-tailed distribution used).