In this paper, a new method for estimating shallow- water depths (bathymetric map) from multispectral images is proposed. This method is based on using Artificial Neural Network fitting algorithms using reflectance of bands influencing water depths and their logarithms for bathymetry detection. An automated method for calibrating the parameters for a Log- Nonlinear inversion model was developed using Levenberg-Marquardt training algorithm. The ANN fitting algorithms using Green and Red bands reflectance and their logarithms was compared with ANN using only Green band reflectance, four SPOT-4 image bands reflectance, and two conventional models (Third order polynomial correlation using the Green band Reflectance and Generalized Linear Model using both Green and Red bands reflectance).