With the large scale integration of wind power, having a deep understanding about the stochastic characteristics can lead to more precise forecasting and better utilization on wind power. Wind power fluctuation has received well attention while the probabilistic characteristics regarding wind power states are still under progress. This paper first presents the definition of wind power states, and then studies the probabilistic descriptions of the duration time within one power state and transition between different power states employing massive measured wind power data. The duration time within one wind power state is revealed to follow inverse Gaussian distribution, which is useful for power system operation. The matrix of the transition probability between wind power states quantifies the degree of transition and shows a ridge pattern. The gradient of the ridge is related to the time scale of interest. This work is supported by National Basic Research Program of China (No. 2012CB215106) and ABB Co., Ltd. Research Project.