In modern engineering designs and analyses, computer models are frequently used. Due to the presence of uncertainties associated with the model inputs and parameters, which are treated as random variables, analysis is feasible if the methods employed do not require excessive computations yet produce reasonably accurate results. Point estimate methods are such schemes that are potentially capable of achieving the goals. Assuming normal distributions to the random variables, three point estimate methods (Rosenblueth's, Harr's, and a modified Harr's method) were evaluated in this paper for different numbers of random variables and different model types. Results of this evaluation indicated that the proposed modified Harr's method yielded comparable, if not better, performance than the other two methods. Also, performance evaluation indicated that additional statistical information, other than the commonly used first two moments, should be incorporated, if available, to enhance the accuracy of uncertainty analysis.