Algal bloom highly impacts the ecological balance of inland lakes. Remote sensing provides real-time and large-scale observations, which plays an increasingly significant role in the monitoring of algal bloom. Various Vegetation Indices (VIs) derived from satellite images have been used to monitor algae. With threshold segmentation of VI, the area of algal bloom can be extracted from images. However, the result of threshold segmentation only reflects the condition of algae when images were generated. Compared to separated VI data obtained at a particular moment of time, temporal spectral VI data contains phonological information of algae, which may be used to evaluate algal bloom more accurately and comprehensively. This study chose MODIS NDVI data of the Lake Taihu from 2001 to 2013, and constructed temporal spectral data for each year. Then, we determined the feature temporal spectra of severe cyanobacteria bloom, moderate cyanobacteria bloom, slight cyanobacteria bloom and aquatic plants, and separated these four kinds of objects using SVM (Support Vector Machine) algorithm, getting the spatial distribution and area of them. In order to compare the results of our method with traditional threshold segmentation method, we chose 8 separated NDVI images from the temporal spectral data of 2007. With the threshold 0.2 and 0.4, cyanobacteria bloom was classified into three degrees: severe cyanobacteria bloom, moderate cyanobacteria bloom, and slight cyanobacteria bloom. By comparison, it showed that our method reflected cyanobacteria bloom more comprehensively, and could distinguish cyanobacteria and aquatic plants using the phonological information provided by NDVI temporal spectra. This study provides important information for monitoring the algal bloom trends and degrees of inland lakes, and temporal spectral method may be used in the forecast of algal bloom in the future.