Distributed clustering is an effect method for solving the problem of clustering data located at different sites. Considering the circumstance that data is horizontally distributed, algorithm LDBDC (local density based distributed clustering) is presented based on the existeding algorithm DBDC (density based distributed clustering), which can easily fit datasets of high dimension and abnormal distribution by adopting ideas such as local density-based clustering and density attractor. Theoretical analysis and experimental results show that algorithm LDBDC outperforms DBDC and SDBDC (scalable density-based distributed clustering) in both clustering quality and efficiency.