Urban Building Energy Modeling (UBEM) is an emerging method for exploring energy efficiency solutions at urban or district scales. More versatile than statistical models, physical bottom-up UBEMs allow planners to quantitatively assess retrofit strategies and energy supply options, leading to more effective policies and management of energy demand. The most common approach for formulating an UBEM involves segmenting a building stock into archetypes, characterizing each type, and validating the model by comparing its output to aggregated measured energy consumption. This paper presents a more detailed methodology for setting up UBEMs while faced with incomplete information about the buildings. The procedure calls for defining unknown or uncertain parameters in archetype descriptions as probability distributions and, if available, using measured energy data to update these distributions by Bayesian calibration. The methodology is validated on residential houses in Cambridge, Massachusetts. Distributions for uncertain parameters are initially generated using a training set of 399 homes with monthly electricity and gas consumption records and then applied to a larger test set of 2263 homes. The procedure is applied both for monthly and annual metered energy usage data. Results show that both annual and monthly Bayesian calibration lead to significantly better annual energy use intensity (EUI) fits compared to traditional deterministic archetype definitions. As expected, an UBEM calibrated with monthly metered data more truthfully mimics monthly EUI distributions than one based on annual data, revealing the benefit of calibrating UBEMs using the smallest measurement time step available. (C) 2016 Elsevier B.V. All rights reserved.