Proper vegetation indices have decisive influences on the precision of hyperspectral estimation models for surface parameters. In the present paper, in order to find the proper hyperspectral indices for cotton canopy water content estimation, two water parameters for cotton canopy water content (EWTcanopy, equivalent water thickness; VWC, vegetation water content) and corresponding hyperspectra data were analyzed. A rigorous search procedure was used to determine the best index predictors of cotton canopy water. In the procedure, all possible ratio indices and normalized difference indices were derived from the canopy hyperspectra, involving all the two-band combinations between 350nm and 2500nm. Then the correlation between two water parameters and all combination indices were analyzed, and the best indices which produced maximum correlation coefficients were determined. Finally, the indices were compared with the published water indices for their performances in estimation of cotton canopy water content. The results showed that for the estimation of EWTcanopy, the new developed ratio index R1 475/R1 424 and normalized difference index (R1 475-R1 424)/(R1 475+R1 424) was the most proper one, and the correlation coefficient of the estimated and measured EWTcanopy reached 0.849. For the estimation of VWC, the performance of published index was better than new developed index, the best suitable water indices for VWC estimation were (R835-R1 650)/(R835+R1 650), and the correlation coefficient of the estimated and measured VWC was 0.849.