In 1987, a fire in a kitchen in Sharon, PA resulted in the deaths of three persons in upstairs bedrooms, one with a reported blood carboxyhemoglobin content of 91%. Considerable physical evidence remained. The fire was successfully simulated at full scale in a fully instrumented seven room test called SHARON 2. The data collected during SHARON 2 have been used to evaluate the precision of two multiroom computer fire codes: FAST 18 and HARVARD 6.3. A coherent ceiling layer flow occurred during the SHARON 2 simulation and quickly carried high concentrations of carbon monoxide (CO) to remote compartments. Such flow is not directly accounted for in either computer code. However, both codes predict well the carbon monoxide buildup in the sixth room (i.e., the room most remote from the fire). Prediction of the preflashover temperature rise was good. Prediction of temperatures after flashover of the room of origin was less successful. Other predictions of conditions throughout the seven test rooms varied from good approximations to significant deviation from test data. Hypotheses are presented as to the reasons for the differences. At least some are believed due to phenomena not considered in the computer codes.