In 1982, the Nuclear Regulatory Commission posed a forced-ventilation problem which is described. The algorithm developed for its solution is discussed. This was embedded in the Harvard Mark 5.3 computer code, and predictions then made for the outcome of aseries of experiments carried out at Lawrence Livermore Laboratory (see previous paper). Comparisons are made between the predicted and the actual r,esults. Predictions are generally in good agreement with experiment, especially for the volumetric input rate and the CO2and oxygen concentrations. The Morton et al. (1956) virtual point-source plume which we use is shown to be adequate, from the correct predictions of O2and CO2concentrations in the layer. One inadequate feature, however, is the model's consistent underprediction of the peak temperatures reached in the upper layer. This is most readily explained by noting that heating of the floor and of the lower layer have been neglected in the model; they will each lead to increased upper layer temperatures. The pyrolysis rate for the polystyrene pool fire was overestimated; the most likely explanation for the difficulty is that radiation blocking (by pyrolysis gases) has been neglected.