For the design of natural ventilation systems for passive cooling in buildings, engineers and architects are interested in the prediction of ventilation rates as a function of position and size of the ventilation openings. In common use, there are both simple and detailed ( i.e., multizone) ventilation models which rely basically on the same Bernoulli algorithm to describe airflow through large openings. An important source of uncertainty is related to the attribution of discharge coefficients. The present study was undertaken to improve our knowledge on velocity and discharge coefficients when measured in real buildings. The experiments were performed on a naturally ventilated three-level office building where the staircase acted as exhaust chimney. In order to keep the flow pattern stable, a condition for air flow measurements to be reproducible, the experiments were performed on windless nights where the flow was only driven by stack pressure. Air flow patterns were traced with smoke and tracer gas. In a first set of experiments, air velocities, contraction and velocity coefficients and the position of the neutral pressure level ( NPL) have been measured, in a second set of experiments, the resulting effective area of a combination of two openings in series. Air flow rates derived from velocity measurements in the open doorways were found to be in agreement with the flow rates obtained with a constant injection tracer gas technique, with an uncertainty of +/-20%. The velocity coefficients phi=0.7 +/- 0.1 and jet contraction coefficients epsilon = 0.85 +/- 0.1 found in the experiments are shown to be in agreement with the generally accepted value of the discharge coefficient C-d = phi epsilon = 0.6 +/- 0.1, giving new justification for its use in the models. Basic configurations for ventilative cooling are given to illustrate how qualitative modeling used in simple models can give valuable information to the designer.