Abstract: A fusion model of sonar and vision data has been established.A new representation has been put forward which is used to describe the under water environment and obtain the fusion map by means of sum of Gaussians.Firstly，assuming the observe information is Gaussians distribution，the sonar and vision data have been transformed and projected to a common state space，and then the other information of sensors has been weighed and added to it，finally a sensory map which is tractable for computer has been built.The algorithm of position estimation and map matching has been proposed.Finding the maximum of the correlation coefficient and updating the position of the robot can obtain the optimum position estimation.The simulation results show the fusion map used to estimate the position of under water vehicle is consistent，computable and effective.