This paper considers the application of soft computing techniques for predictive modelling in the built sector and presents the extension of the results from previous works of the author. While the latter considers only short-term modelling which is used mainly for control purposes, the present paper discusses also long-term modelling results that may be used for efficiency evaluation in buildings. Three different types of buildings are considered, an air-conditioned zone, a naturally ventilated room, and an endothermic building. The are subjected to their normal occupancy effects and the natural external climatic disturbances which are difficult to incorporate in accurate modelling using conventional quantitative methods. The approach adopted here uses fuzzy logic for modelling, as well as neural networks and genetic algorithms for adaptation and optimisation of the fuzzy model. Takagi-Sugeno fuzzy models are built by subtractive clustering to provide initial values of the antecedent non-linear membership functions parameters and the consequent linear algebraic equations coefficients. A method of extensive searching the possible solution space is presented which explores all the possible permutations for a specified range of orders to derive the initial fuzzy model. This model is an extension of the traditional ARMAX (Auto Regressive Moving Average Exogenous) model where the effect of the moving average term has been accounted for by the fuzziness and its ability to represent uncertainty. The fuzzy model parameters are further adjusted by a back-propagation neural network and a real-valued genetic algorithm in order to obtain a better fit to the measured data. Model validation results using data from the three buildings are presented where the initial (fuzzy) and the improved (fuzzy-neuro and fuzzy-genetic) models are compared and analysed with each other and with conventional (non-fuzzy) models.