The invention discloses a traffic cell division method based on a multi-path spectral clustering theory. The method comprises: step one, basic data of road network division are obtained; step two, according to locations of road detectors, road network nodes are numbered; step three, a weight value of a road section between each two adjacent nodes is calculated; step four, a road network density Laplacian matrix is established; step five, a feature value and a feature vector of the Laplacian matrix are calculated; step six, non-zero first k minimum feature values in the Laplacian matrix and corresponding feature vectors are selected and clustering is carried out by using a k-means clustering algorithm to form a K cluster; step seven, traffic flow densities of all road sections in all areasare extracted; step eight, whether traffic flow density mean value variance checking among all areas is obvious is checked; and if not, the step six is carried out again, wherein the K is equal to K+1; step nine, the areas are divided by boundary intersections; and step ten, division is completed. Therefore, the process of division is simplified; and an objective of minimization of regional internal road network differences and maximization of road network characteristic differences among areas is achieved.