Wells drilled in the core area of shale reservoir are predominantly horizontal compared to vertical wells in conventional reservoirs. Due to the different features of data in horizontal and vertical wells, the 3-D property modeling of shale reservoirs is distinct from the conventional reservoirs. Lateral drilling usually leads to the repeated sampling in the specific vertical intervals and the missing sampling in the bottom part of the shale reservoir. This biased sampling distorts the distribution characteristics in well data from the real reservoir properties. Moreover, because of the large inclination angle in the laterals, it is so common to penetrate only the upper or lower part of the grid. Given that the same method is utilized to assign well values to these partially penetrated girds and fully penetrated grids, the assigned value in these partially penetrated grids fail to represent the reservoir property in the associated grid layer. Therefore, we developed a grid filtering method to solve the issues arised from the biased sampling and the partially penetrated grids. It includes five parts: (1) divide wellbore into multiple well segments using formation top data; (2) recognize the type of well segments and well-penetrated grids to generate a grid filter; (3) process the log-format data of each complete sampling segment using the grid filter; (4) generate one dataset for analyzing distribution features and another dataset as the control points of geostatistic simulation algorithms; (5) build the 3-D property models using the analyzed distribution features and the control points as the input. By generating two independent datasets, we achieved the maximum utilization of log-format data in horizontal wells and eliminated the biased sampling effect. The total organic carbon content and shale lithofacies in the JY-1 district, Jiaoshiba Area, Eastern Sichuan Basin, were used to demonstrate our method.