Cyclic spectrum density estimation plays a significant role in cyclostationary signals analysis. Generally, the methods of cyclic spectrum density estimation are mainly focus in time-domain, such as FFT Accumulation Method (FAM) and Strip Spectral Correlation Algorithm (SSCA). In this paper, based on the principle of frequency smoothing methods, an improved method in frequency-domain was proposed, which could reduce the computational cost greatly by using recursion. The recursive method was based on the fact that the range difference of smoothing window was small when window slid. In other words, we could get a new point of cyclic spectrum density just by computing the complex conjugate products of newly coming into window. In addition, the simulation results were given at last, which showed the good performance of our proposed method.