Abstract The cost optimal method (COM) as applied in the Energy Performance of Buildings Directive (EPBD) uses “non-calibrated deterministic reference buildings (RBs)”. Such RBs are defined with single envelope and equipment parameter values, for which calibration with actual building stock energy performance (EP) is not undertaken. Thus, it is not possible to visualise the effect of uncertainties or diversity in the input parameters on cost-optimal level benchmarks and to verify the choice of RBs. The paper proposes an update to the COM via use of “Probabilistic Bayesian calibrated RBs” to handle uncertainties and produce more realistic cost optimal levels to support policy makers in devising effective fiscal and legal support mechanisms while facilitating harmonised EP benchmarking between MS for different building categories. The process is validated by findings from Urban Building Energy Modelling studies showing that Energy Use Intensity (EUI) distributions from “probabilistic Bayesian calibrated RBs” versus “deterministic RBs” match significantly closer to the measured EUI distributions with a resulting reduction in the Kolmogorov–Smirnov (KS) test results of up to 82% when monthly calibration was performed.