The salinity data in the ocean are non-uniform and irregular, therefore methods for salinity estimation using available predictors (i.e. temperature data or others) are mandatory. A set of regression models were presented for estimating salinity profiles in the upper 500m of the Red Sea from the measurements of temperature profiles, surface salinity, and some other predictors. Both temperature and surface salinity were used to capture the curvature seen in temperature salinity plots, latitude and longitude were used to capture systematic spatial variations over the fitting regions and Julian day was used to capture seasonal variability in the region. Hence, the best-fit regression curve and the minimal errors of the salinity estimates were found for the study area. This regression model over all for the entire region at all depths was quadratic in temperature and linear in surface salinity, longitude, latitude and day of the year. Even without the surface salinity measurement we could estimate the salinity with good reduction of RMS errors for all depths below 150m.