Accurate power system state estimation is essential for power system control, optimization, and security analysis. In this work, a model-free approach was proposed for power system static state estimation based on conditional Generative Adversarial Networks (GANs). Comparing with conventional state estimation approach, i.e., Weighted Least Square (WLS), any appropriate knowledge of system model is not required in the proposed method. Without knowing the specific model, the GANs can learn the inherent physics of underlying state variables purely relying on historic samples. Once the model has been well trained, it can generate the corresponding estimated system state given the system raw measurements. Particularly, the raw measurements are sometimes characterized by incompletion and corruption, which gives rise to significant challenges for conventional analytic methods..The case study on IEEE 9-bus system validates the effectiveness of the proposed approach.