We investigated how the degree of correspondence of cross-spectral density matrix of generated inflow turbulence with the target cross-spectral density matrix estimated by the wind tunnel test influenced on LES (Large Eddy Simulation) results of a turbulent boundary layer. Some types of inflow turbulence were generated by a step-by-step method based on Monte-Carlo simulation considering several reproduction levels of target cross-spectral density matrices. LES was conducted using the above types of inflow turbulence. The, results were as follows. The reproduction of cross-spectra of the normal components (u_1-u_1, u_2-u_2, u_3-u_3) in the inflow turbulence generation was very important to reproduce power spectra, spatial correlation and turbulence statistics of wind velocity profiles in LES. When cross-spectra of the shear component (u_1-u_3) were included in addition to those of the normal components, the spatial correlation of the shear component just behind the inflow boundary was more reproducible. However, there were no differences in other regions between the LES results with the cross-spectra of u_1-u_1, u_2-u_2, u_3-u_3 components and those with u_1-u_1, u_2-u_2, u_3-u_3 and u_1-u_3 components, and the target turbulence quantities of grid scale components were reproduced well in each case.
