Abstract: In the field of the analysis of a stochastic system and structural reliability, it is a frequent problem for evaluating a statistical moment of a function of multiple variables. And the point estimate method (PEM) is the simplest and most efficient approach. In order to apply PEM more rationally in a practical engineering system, the computational performance of several typical PEMs is discussed by systematical case studies in detail. And the evaluation precisions for statistical moments, which are for second polynomial and compound one with different combinations of random variables, are obtained by typical PEMs. By comparing these results, it can be found that: 1) for all of PEMs, the precision for higher order statistical moments is lower than the one for lower order statistical moments | 2) the nonlinearity of the function, probabilistic category and coefficient of variation (COV) of random variables influence the precision of PEMs obviously, but the influences of the number and coefficient of the correlation of variables on precision is different for different methods | 3) PEM proposed by Zhao & Ono is the best one among five typical methods, but it is not accurate enough for a function with strong nonlinearity and variables with big COV | 4) the precision of the method proposed by Harr varies drastically when the correlation coefficient is equal to zero.