In this paper, we propose a discord discovery method which finds the k-most dissimilar subgraphs of size n among the subgraphs of the same size of an input graph, where the values of k and n are given by the user. Our algorithm SD3 (Subgraph Discord Detector based on Dissimilarity) exploits a dynamic index structure and its effectiveness is demonstrated through experiments using graph data in chemical-informatics and bioinformatics.