To remove the large-scale and dense noise from the terrestrial laser scanning data and keep the edge features of buildings, a filtering method fusing intensity with density of points is proposed based on the varied distance of the points to the scanning stations. The spatial distribution of noise and the intensity distribution of point clouds are analyzed comprehensively. The spatial quadtree index is established based on the horizontal and vertical angles. The fast clustering of local points and the removal of isolated points are realized based on the account of the distance before and after points in the leaf nodes, and the large-scale and dense noise is filtered out according to the ratio among different types of intensity point numbers in the same point set. The research results show that the proposed method can be used to effectively filter out the large-scale and dense noise involved in the terrestrial laser scanning data with an accuracy of above 90%.