Clustering is an important tool for data analysis and many clustering techniques have been proposed over the past years. Among them are density-based clustering methods, which have several benefits such as the number of clusters is not required before carrying out clustering; the detected clusters can be represented in an arbitrary shape and outliers can be detected and removed. Recently, the density-based algorithms were extended with the fuzzy set theory, which has made these algorithm more robust. However, the density-based clustering algorithms usually require a time complexity ofO(n2) wherenis the number of data in the data set, implying that they are not suitable to work with large scale data sets. In this paper, a novel clustering algorithm called landmark fuzzy neighborhood DBSCAN (landmark FN-DBSCAN) is proposed. The concept, landmark, is used to represent a subset of the input data set which makes the algorithm efficient on large scale data sets. We give a theoretical analysis on time complexity and space complexity, which shows both of them are linear to the size of the data set. The experiments show that the landmark FN-DBSCAN is much faster than FN-DBSCAN and provides a very good quality of clustering.