This paper presents a new method for the localization of multiple narrowband acoustic sources using spherical microphone arrays. The space-time array signal model is transformed into spherical harmonic domain by spherical Fourier transform (SFT). Then subspace decomposition technique is used to obtain signal and noise subspace in the domain. Relying on the recursive relationship of spherical harmonics, signal subspace can possess the displacement invariances which could be used to estimate both the azimuth and elevation angles of acoustic sources impinging on spherical arrays. Simulations demonstrated that this algorithm has a high accuracy of estimating multiple acoustic sources with low computational complexity. Moreover, spherical microphone arrays have the advantage of spherical symmetrical structure, facilitating 3-D direction-of-arrival (DOA) estimation, which is superior to uniform circular arrays (UCA).