The accuracy of zero cross time detection is essential to the phase and frequency measurement of 50Hz AC signals. The most commonly used zero cross time detection methods are for synchronization check which requires low accuracy. A new method for zero cross time detection of AC signal is proposed. The peak detection method and Levenberg-Marquardt algorithm are used to estimate the parameters of sinusoidal model based on the continuous sampled data. The zero cross time is calculated according to the model and estimated parameters. This method and test bench is established on MATLAB. Experiments on different sample-rates, data length, harmonic power, and signal-noise ratio were conducted to verify the performance of this method. The results indicate that the deviation of detected zero cross time is around 1μs in the circumstance of sample rate of 100ksps, data length of 104, signal-noise ratio of 40dB minimum, and harmonic power of 1% maximum.