The covariance matrix, employed for measuring the linear correlation between variables, plays a vital role in data analysis, such as statistical prediction and hypothesis testing. When the data dimension is high, the traditional sample covariance matrix is not an ideal estimator of the population covariance matrix anymore, resulting in degradation or even inaccuracy of the second-order moment estimators' performance based on the sample covariance matrix. This paper studies the unbiased estimators of the second-order moments under complex Gaussian distribution. The proposed unbiased estimators have better statistical properties and numerical performance than the existing estimation methods.