The paper presents a comparison of simulation results obtained by using the American Thermal Analysis Research Program (TARP) with data measured in four zones of an unoccupied real building, subjected to various heating patterns during three months. The main findings are: general discrepancies in internal air, surface and mean radiant temperatures are small(+/-0.5 degrees C), but some systematic deviations are observed, e.g., actual decreases in air temperature, upon shutdown of heaters, are up to 1.5 degrees C smaller than the simulated values, indicating that the program underestimates the overall thermal inertia; and actual decreases in internal surface temperature of external walls are larger than the simulated values, indicating that the program overestimates the external walls＇ thermal inertia. The simulation of general power demands is also adequate (+/-14%), but the estimate of maximum power demand (during the first hour of heating) is very poor (up to -50%), indicating the above-mentioned underestimate of overall thermal inertia. Total energy estimates for long (months) and short (weeks) periods are excellent (+/- 0.95% and better) for cold and cool winter periods, but much poorer (up to +/-14%) for very warm winter periods. The paper discusses the possible flaws and wrong assumptions in input data, and shows that, with careful choice of this data so that the problem at hand is well approximated, it is not the data that causes the systematic discrepancies but an inherent inaccuracy in TARP in the mathematical-physical modelling of thermal inertias.