The estimation of pulsar period is important for pulsar searching, observation and timing. In this paper, the frequency-domain method which is popular in periodicity searches is analyzed, and further more, a novel method in time-domain is proposed to estimate the cyclo-period of pulsars based on the second-order cyclo-stationary model of the radiation signal. Firstly, the discrete-time observing signal is blocked by assumed period which is alterable. Then, the correlation estimations and their sample variances are computed each as a function of blocking operator index, where the largest peaks display a periodic pattern, from which the cyclo-period can be obtained. Using this method, the period of PSR J0437-4715 with single pulse is estimated well based on original signal, and the period of PSR B1821-24 with double pulses is also estimated based on the simulated signal. The same work is done for the observed data from ATNF EPN pulsar database. The simulation results and the experimental results for many pulsars show that compared with the classical Fourier-transform method, this algorithm is non-sensitive to the noise, and it is effective even though in the case of low signal-to-noise ratio. And it is of low computation complexity compared with the bispectrum coherence method. This method possesses higher popularization value in real-time observation and processing for weak pulsar signal, though it does not work well in dealing with noncontinuous observing data.