Resumo: A technique for the estimation of the phase function in natural waters, based on the least squares estimator, which is associated to a regularization function (maximum entropy principle), is presented. In the analyses, we utilized sinthetic data (radiances and irradiances) corrupted with noise. These data were generated by the same analitical forward model utilized in the inversion and implemented by \textit{Peesnc} code. The inverse problem is iteratively solved by the quasi-newtonian optimization algorithm implemented in the \textit{E04ucf} routine of the NAG Fortran library. The recovery of the phase function is successfuly performed for the selected sample problem.