To retrieve chlorophyll a (Chla) concentration in Taihu Lake by three-band model, a field study was conducted on August 19, 2004 to collect water samples (N=38), which contained widely variable Chla (7.8-154.3 microg x L(-1)) and total suspended solids (65.0-190.2 mg x L(-1) dry wt), and the synchronous Hyperion images was also acquired as remote sensing data. After obtaining the approximate range of wavelengths for the three bands by analyzing the inherent optical properties of Taihu Lake, the three-band models were spectrally tuned to select the bands for most accurate Chla estimation. Finally Hyperion B34 (691.37 nm), B37 (721.90 nm) and B50 (854.18 nm) were selected to establish a three-band model. The results show that strong linear relationship is found between analytically measured Chla and the three-band model (r = 0.934), which accounts for 87.2% of variation in Chla and allows estimation of Chla with a root mean square error (RMSE) of 13.93 microg x L(-1), whereas the traditional two-band models accounts for lower accuracies of Chla estimation (spectral ratio, R2 = 0.844, RMSE = 15.41 microg x L(-1), and reflectance first-derivative, R2 = 0.831, RMSE = 16.00 microg x L(-1)). The findings prove that the three-band model is applicable for Chla retrieval in turbid, productive inland waters and by using Hyperion hyperspectral data.