Head-Related Transfer Function (HRTFS) is the key to many applications in spatial audio. Its large amount of data makes it difficult to make real-time implementation. Reducing HRTF data is necessary and important. In this paper, we apply a new developed signal decomposition theory, named Adaptive Fourier Decomposition (AFD), to decompose and compress HRTF data, comparing with traditional Fourier's convergence property and PCA's compression property. Simulation results show that the proposed AFD-based decomposition and compression method enables evident performance improvement for HRTF.