Abstract: An accurate array manifold is necessary for array signal processing.However,it is difficult to measure the positions of the horizontal array elements on ocean bottom.Extracting the empirical Green's function from ambient noise provides an alternative to calculate the array element positions passively.The distance between two array elements can be obtained from the superimposed cross-correlation functions.One way is to build the nonlinear overdetermined equations,and solve the question by the Least Square(LS) method.This paper uses a method called MultiDimensional Scaling(MDS) to deal with this issue.It constructs the Gram matrix of position vector,and acquire the array shape by the eigenvalue decomposition of Gram matrix.Numerical simulations and experimental results verify the effectiveness of MDS and show that MDS is more robust than LS.