The parameter uncertainty in a physics-based building energy simulation model is caused by lack of information or incomplete knowledge. To bridge thegap, the model is generally calibrated by estimating uncertain parameters. Parameter estimation methods can be classified into (1) trial and error, (2)sampling-based, (3) optimization-based, and (4) stochastic approach, e.g., Bayesian calibration. Recently, Bayesian calibration has been widely used forparameter estimation problems in the building energy model. However, the accuracy of Bayesian calibration is influenced by (1) prior distribution assumedby expert’s subjective knowledge, (2) likelihood function for Gaussian distribution often leads to the parameter identifiability problem and low accurateestimation results, and (3) posterior distribution approximation method such as Markov Chain Monte Carlo that demands significant computation time.A new approach that uses Generator Regularized continuous conditional Generative Adversarial Networks (GRcGAN) is presented in this paper toovercome the Bayesian calibration. GRcGAN is well-suited to solve the inverse problem with continuous conditions. In the paper, ten uncertain parametersof the DOE reference building model were selected. Then, 1,200 sets of ten parameters were sampled using Sobol sequence and used for EnergyPlussimulation to calculate monthly electricity and gas uses. It is shown in the paper that the GRcGAN model can successfully estimate ten uncertain parametersbased on monthly electricity and gas uses. The CVRMSE of the calibrated EnergyPlus model is less than 2% after 3,500 epochs. The proposed approachwill be beneficially used for model calibration as well as for estimating unknown parameters by a building energy analyst.