The LUTCHI data are the main colour appearance data used as the basis of many colour appearance models including CIECAM97s. It was shown in the LUTCHI data that projected colours are very different from reflective colours however there are relatively fewer data for projected colours than for reflective colours. In this study, it is intended to expand the colour appearance data of projected and self-luminous colours. The additional colours would then help investigate the performance of existing colour appearance models and, if necessary, enable the derivation of a new model to improve performance for projected and self-luminous colours. Before the colour appearance study, firstly the performances of the instruments and the displays used in the study were investigated. It was found that LCD displays perform very differently from CRT monitors. Two mathematical characterisation models for LCD displays were developed named S-Curve Model I and S-Curve Model II. The new colour appearance data set, CII-Kwak, was accumulated by a series of psychophysical experiments. The magnitude estimation technique was applied with the same experimental set-up as for LUTCHI experiments. The CII-Kwak data set has 20 phases with a total of 28,608 estimations covering various displays, luminance of a reference white, background luminance factors, surround conditions and stimulus sizes. Based on the CII-Kwak and the LUTCHI data set, the colour appearance phenomena were analysed. It was found that there are systematic colour appearance changes by the viewing factors investigated. Also eight colour appearance models were tested using the CII-Kwak and LUTCHI data sets. CIECAM97s-based models performed similarly well, but all models tested failed to predict several colour appearance changes, especially under dark surround conditions, which lead to suggest a new colour appearance model to have a better performance for colour appearance predictions. The Kwak03 model was derived from the CIECAM02 with several major modifications such as the cone signal ratios and the omission of the dynamic function. The Kwak03 model was shown to outperform all the other colour appearance models tested and also to be capable of predicting all colour appearance phenomena found in this study with good accuracy.