Empirical validation of building energy simulation tools is an important component in assessing the reliability of the simulation software. An experiment performed in conjunction with the International Energy Agency's Task 34/Annex 43 was used to assess the performance of four building energy simulation codes used to model an outdoor test cell with a glazing unit. The experiment was run for a 20-day period during October 2004, and experimental cooling powers were compared with predictions from (1) EnergyPlus, (2) DOE-2.1E, (3) TRNSYS-TUD, and (4) ESP-r. Detailed code inputs for optical and thermophysical properties as well as the impact of thermal bridges were quantified through experiments and simulations; numerous statistical parameters and sensitivity analyses were implemented to facilitate a thorough comparison of predicted and experimental cooling powers. The mean percentage differences for all four codes were: 1.9% for EnergyPlus, −3.6% for DOE-2.1E, −6.2% for TRNSYS-TUD, and 3.1% for ESP-r. The implications of various modeling procedures as well as a detailed discussion of the results are provided, specifically concerning the sensitivity of the code cooling power predictions to the selection of convective heat transfer coefficients and algorithms.