The first and significant step for processing airborne LiDAR is to remove non-terrain point clouds and reserve ground point clouds. According to the irregularities of elevation on LiDAR point clouds in the spatial distribution, a window iterative Kriging algorithm is proposed for filtering off objects from terrain point clouds. First of all, an elevation histogram of point clouds is used to filter low and high outliers. Then average point spacing is taken as the size of initial window, a Kriging interpolation method is adopted to fit the elevation of central grid by using the elevations of eight neighbor grids. If the height difference between fitting value and original height is larger than a height threshold, point clouds lying in the grid cell would be classified as object points. Then surplus points are interpolated into new grids with a size of window is twice as big as previous one. With the exponential increase of the size of window, surplus point clouds continue to be classified until the biggest window size is reached. Fifteen sampled data provided by ISPRS is used to test the method and eight other algorithms are compared with this method. The results show that type I error and total error of the method are less than the corresponding errors of most other methods. Therefore, the algorithm has some reference values for filtering LiDAR point clouds.