In Cognitive Wireless Sensor Network (C-WSN), spectrum utilization and energy-efficiency are both significant items for the whole network. Interference between Primary User (PU) sensor and Secondary User (SU) sensor should be eliminated in order to realize spectrum sharing. In this paper, mathematical model of multi-carrier power allocation in cognitive OFDM is constructed. Multi-carrier power allocation based on rate adaptive criterion is proposed under the constraints of SUs’ power control. An efficient subcarrier power allocation algorithm based on adaptive water-filling is proposed. The improved algorithm could directly determine the sub-carriers that do not require additional power injection by rough estimation of water levels. Computational complexity of proposed algorithm could reduce rapidly. Meanwhile, theoretical derivation and numerical results both indicate that, with the proposed power allocation algorithm, SU’s spectral efficiency is superior to other traditional water-filling schemes, and the algorithm also has some adaptive features for practical implementation in C-WSN.