It is very important to determine the neighbor relationship of the monitoring points in Internet of Things when abnormal monitoring data is audited. Traditionally the neighbor relationship is directly set according to administrative area or geographical location of monitoring points, whereas some shortcomings need to be overcome. We propose a method for determining the neighbor relationship of the monitoring points in Internet of Things based on historical monitoring data. It adopts the clustering analysis method, and chooses cluster number and clustering algorithm according to silhouette coefficients. Experiments on real monitoring data show that the neighbor relationship obtained through the proposed method is more in line with the true situation, and can provide a more scientific and reasonable basis for verifying the validity of the monitoring data of Internet of Things.