This study aimed to investigate the performance of genetic algorithms coupled with partial least squares (GA-PLS) modeling of spectral reflectance in retrieving equivalent water thickness (EWT) at leaf and canopy level. A genetic algorithm was used to identify a subset of spectral bands sensitive to the variation in EWT, and PLS was then applied to relate the identified bands to EWT values. GA-PLS was applied to leaf level reflectance available from LOPEX dataset, and canopy data includes reflectance simulated by a leaf radiative transfer model PROSPECT and a canopy radiative transfer model SAILH and acquired by airborne visible/infrared imaging spectrometer (AVIRIS). The results indicate that GA-PLS has the capability of retrieving EWT from leaf and canopy reflectance, and achieved good estimation accuracy, i.e. low root mean square errors (RMSE) and high squared correlation coefficients (R2). For the retrieval at leaf level, the estimation accuracy can be as good as RMSE = 0.0019 g/cm2 and R2 = 0.939 or better. For the retrieval at canopy level, the model accuracy is RMSE = 0.0061 g/cm2 and R2 = 0.966 or better when PROSPECT-SAILH simulated reflectance was used; when AVIRIS image spectra were used, the model accuracy is RMSE = 0.0094 g/cm2 and R2 = 0.8734 for the calibration, and RMSE = 0.0132 g/cm2 and R2 = 0.7756 for the validation. These results from GA-PLS modeling support the conclusion that GA-PLS has the potential to be applied to AVIRIS, Hyperion and HyMap imagery for retrieving EWT. The selected bands for the AVIRIS datasets differ from those for the LOPEX and PROSPECT-SAILH simulated datasets, and this inconsistence of the selected bands for different datasets indicates that the GA-PLS method has the advantage of tuning the optimum bands for PLS regression and accommodating the effects of confounding factors.