In fields such as healthcare, data scarcity severely hinders deep analysis. Learning the underlying data dependencies via neural networks and generating similar data accordingly is an effective approach to mitigating data scarcity. However, current data generation techniques suffer from problems such as mode collapse and the limited ability to fit complex data distributions. At the same time, some of the generated data are untrustworthy, leading to uncertainties in downstream tasks. To solve these problems, this paper introduces a trusted tabular data generation architecture: TrustGenTable. In this scheme, we proposed a novel method to capture complex distribution in tabular data. To find trusted generated data, we proposed a checking mechanism that conducts both specific and adaptive checking. Extensive experiments on real-world datasets show that our proposed architecture can effectively generate trusted data and significantly improves the impact of data scarcity on downstream tasks.