Abstract: The complexity of a forest community is defined as having the average amount of its information by eliminating the uncertainties of species and layers of a tree individual randomly selected from all trees in the forest community. The joint entropy H(X,Y) is proposed to measure the complexity of a forest community H(X,Y)=H(X)+H(Y|X), in which H(X)=-∑[DD(]S[]i=1[DD)][SX(]ni[]N[SX)]log2([SX(]ni[]N[SX)]) and H(Y|X)=-∑[DD(]S[]i=1[DD)][SX(]ni[]N[SX)]∑[DD(]4[]j=1[DD)][SX(]nij[]ni[SX)]log2([SX(]nij[]ni[SX)]), where S stands for the number of tree species (X), N for the total number of individuals in the forest community, ni for the number of the ith tree species, and nij ?for the number of the ith tree species in the jth layer. H(X) is defined as the compositional complexity of tree species and H(Y|X) as the structural complexity of tree species. The higher the H(X,Y) value, the greater the complexity in the forest community. A case study is presented based on the survey data from three types of forest communities in Heishiding Nature Reserve, Guangdong Province. Three sampling plots were established, each with a size of 60 m×60 m, representing coniferous forest, mixed coniferous broadleaved forest and evergreen broadleaved forest. Each plot was divided into 36 quadrats with a size of 10 m×10 m. The data for all trees with DBH1 cm were gathered, including their coordinates in the sampling plots. Tree sizes were divided into four categories based on their DBH: DBH1, 5, 10, and 30 cm. Using computer simulation, 13 types of quadrat sizes (12 m×12 m, 16 m×16 m, …, 60 m×60 m) within a plot were objectively selected based on the method of nested quadrat sampling. The results show that the order of H(X,Y) of three typical forest types is as follows: evergreen broadleaved forest > mixed coniferous broadleaved forest > coniferous forest. At the same time, the fractal relationships between H(X,Y) and sampling size among the three forest types reveal that H(X,Y) has a statistical selfsimilarity feature.