Uncertainty Quantification (UQ) employing a Monte Carlo Sampling (MCS) method in a building simulation domain has been widely used to account for risks of predicted outputs for robust decision making. However, the stochastic approach for UQ problems requires significant computational burdens compared to the deterministic approach. This paper addresses two surrogate models (Gaussian Process Emulator (GPE) and Polynomial Chaos Expansion (PCE)) which together can be regarded as a meta-model of a Building Performance Simulation (BPS) tool with a high-fidelity model. In the paper, the developed GPE and PCE with different model structures were compared in terms of a prediction capability under different amount of training data and number of inputs. The aim of the comparative study is to identify the relative prediction abilities and model flexibility of GPE and PCE. It was found that the GPE and PCE produce high performance qualities having fast computation speed compared to the developed basis model if new inputs having identical inputs and probability ranges were used. In terms of two-sample Kolmogorov-Smirnov (K-S) hypothesis test, mean values of the minimum p-values of the GPE and PCE were 0.999 and 0.569, respectively, if the number of samplings are over 30 cases. Otherwise, the PCE shows significantly reduced performance quality than the GPE.