The objectives of the current study were to investigate the opportunity of estimating soil salinity from hyperspectral data and identifying the most informative spectral zones for estimation. Electrical conductivity (EC) measurements of ninety topsoil samples (0–30 cm) collected from Toshka, Egypt, were used as data set. Analytical spectral device was employed to collect the reflectance spectral signatures of soil samples. Both linear regression and HSD Tukey's analyses displayed that the SWIR1 and SWIR2 zones are the most suitable for soil salinity prediction while, blue, green and NIR were the wickedest. Moreover, EC estimation was better in case of lower soil salinity (0-2 dS m^-1) than higher levels (8-¹). Partial-least-squares-regression (ΡLSR) was employed to establish soil salinity prediction model using the training set of soil samples (n=75). The PLSR model was set up using the most informative wave bands (SWIR1 and SWIR2). The result showed that PLSR linear model gave a precise prediction of soil salinity (R² = 0.93). The results revealed that employing reflectance values in SWIR in the model variables increases the precision of soil EC prediction.