Large discrepancies can occur between building energy performance simulation (BEPS) outputs and the actual building energy performance. Uncertainty and sensitivity analysis are performed to discover significant contributions of each input parameter to these discrepancies. Variance-based sensitivity analyses typically require many stochastic simulations which is computationally demanding. To overcome these impediments, this study proposes a reliable meta-model-based sensitivity analysis, including validation, Morris’ method, meta-modeling and Sobol’method, to identify the most influential input parameters on BEPS prediction (annual energy consumption) at the early building design process. Four meta-models (Multivariate Adaptive Regression Splines (MARS), Polynomial Regression, Random Forest and Support Vector Regression) are evaluated to compare their performance using statistical metrics. The comparison shows that MARS achieves the best performance. A hypothetical building is used to analyze the proposed methodology. It is concluded that the cooling set-point temperature and g-value of the window are the most influential input parameters for the analyzed case study.