To validate zone models from experimental data, the experimental data needs to be reduced to a form which is compatible with the zone model. Two parameters in zone modelling which experimental data needs to be reduced to before comparison, are interface height and the upper and lower zone temperatures. Fire experiments of three different fire sizes were housed in a full sized double room enclosure. Thermocouple trees were located in the comer and centreline of each compartment. During the analysis of the experimental data three interface height prediction methods were used. A commonly used empirical method known as the N% method did not perform well in the fire compartment. The maximum slope method which estimates the interface height as the point where the temperature change is maximal over height, worked very well for the data reduction as did Quintiere's method, which used two integral identities to solve for the interface height. The interface height determination for these methods could be used successfully with temperature averaging techniques however would not be sufficient to validate the zone models interface height calculation. The interface height does not represent any physical occurrence, rather a layer of mixed gases appears in the room between the zones. To be compared conservatively to a zone model the height at the bottom of this interface layer should be used. To determine average temperatures the interface height directly in the middle of the interface layer would be best. Six temperature averaging techniques were investigated. All predicted the lower zone temperature accurately. Quintiere's method was the most successful in accurately predicting the upper layer temperature, as it was not affected by thermocouple readings in the interface layer. Emmon's method ranged from slightly over predicting the upper zone temperature to greatly over predicting it. Spatial averaging, Averaging based on the equation of state and Janssens and Trans method all slightly underestimated the upper layer temperature. These methods work more successfully if averaging takes place not from the boundary specified at the interface height but from the height above and below the interface layer.