This thesis describes the design process and analysis of a multichannel virtual sound reproduction system. The objective is to find the optimal configurations that produce virtual sound field as close to real one as possible. Since impulse responses at the surface of a rigid sphere show characteristics similar to those of real HRTFs, rigid sphere model is selected to model the sound propagation paths in scattered fields of a system. Evaluation of the system performance was done by introducing three factors: accuracy of signals with given system resources, energy required in multichannel signal processing, and robustness of sound image. Two-channel systems and four-channel systems were studied through computer simulations. The results can be summarized as follows: First, as the in-between angle of loudspeakers decreases discrepancy between real field and virtual field also decreases. Second, when the in-between angle of loudspeakers decreases, the energy required in signal processing increases and the output signals are more easily distorted. In the case of very closely spaced configurations, multichannel driver controls the pair of loudspeakers so that they behave like a dipole. Finally, robustness of a sound image was evaluated by calculating the area of zones of equalization. It turned out that narrow configuration is desirable for increasing the robustness of sound images. With the criteria suggested, an optimal configuration for a given situation can be determined by finding a compromise between them.