Deep Forest is a highly competitive algorithm compared with deep neural networks. However, it suffers from large amounts of calculation due to the large forest quantity. This paper proposes the first hardware accelerator for Deep Forest based on FPGA. Firstly, a delicate node computing unit (NCU) is designed to improve inference speed. Secondly, an efficient architecture and an adaptive dataflow are proposed to alleviate the problem of node computing imbalance. The experimental results show that the proposed design can achieve at least 40x speedup on Intel Stratix V FPGA compared to that on a 40 cores high performance x86 CPU.