Let a 2D random function X(x,y) to denote fBm with exponent 0 < H < 1, then its spectral density Sx(u,v) has relation: Sx(u,v) 1/(u2 + v2)H+1. Such algorithm based on fBm has shown us beautiful pictures of fractal mountains. But the mountains (fractal surfaces) were produced naturally by random process. As a result, the macroscopic shapes and global positions of fractal mounts are not controllable. This paper presents a method that generates fractal mountains with controllable macroscopic shapes and positions using spectral synthesis. First, the discrete data of Y(x,y) on finite grids are inputted, and FFT is employed to produce discrete spectral F(u,v). Second, by InvFFT, low frequency components of F(u,v) together with high frequency components of F(u,v) are transformed to produce Z(x,y)--fractal surface. The macroscopic shapes are controlled by low frequency; meanwhile, the high frequency describes texture of fractal mountains.