Abstract. The wind energy production estimates are very important to a wind power project. And, the remote sensing technique has been widely used to obtain the offshore wind speed and direction which could be used to calculate the wind energy of potential wind farm. However, the directional wind energy distributions are rarely studied, which also play important roles in analysis of wind farms' potential power. In this article, the minimum number of records to obtain offshore directional wind distribution is stated by simulation experiment on In-situ dataset. The NDBC buoy dataset is randomly and multiply sampled to build new dataset under different numbers of observation records, which vary from 21 to 800. The resample under the same number of observation is repeated for 100 times to build dataset group. The directional wind distribution of new dataset is compared with the one of original buoy dataset, and errors made by dataset with fewer records are calculated. Besides, the 10th largest error in the sampled dataset group, which have the same number of observation records, is regarded as the error bound for those dataset. The change rule of the error bound is shown by fitted curves. Based on the fitted curves, minimum number of records is calculated. By this simulation experiment, the minimum number of records to represent wind direction frequency is 350, and 800 for annual direction distributions of wind energy density. To reduce the number of records needed in retrieval, some methods are discussed and tested.