Computer models used to predict the response of reactors to hypothetical accidents necessarily incorporate approximating assumptions. When attempting to verify the models by comparing predicted and measured responses in test facilities, there must be confirmation that these assumptions are realistic. Recent experiments in U.S. facilities capable of repeatedly duplicating the transient behavior of a pressurized water reactor undergoing a pipe rupture show that the assumption of complete water-steam mixing during the transient results in the predicted decompression being faster than that observed. A noninstantaneous condensation model is described that will be incorporated into later versions of the RELAP4 code, allowing more realistic predictions of decompression rate during the emergency core coolant injection phases of a loss-of-coolant accident.