Under urbanisation, transport infrastructures may be improved when urban population grows. Meanwhile, land use patterns may vary and this urban dynamics may drive variations in mode choice of commuters and spatial features of transport networks. Empirical studies have observed scaling laws between the amount of transport infrastructures and city sizes. This thesis is aiming to provide a modelling framework for the analytical investigation of network growth and present some empirical observations of the variation in spatial features of transport networks. First, a simple linear monocentric city model is formulated and the global performance of transport systems is derived. Two cases according to strategies of urban intensification and sprawl have been studied to examine the consequence of the scaling-law growth in transport infrastructures. Second, this thesis proposes a modelling framework. The framework includes two congestible modes, the scaling-law growth of transport infrastructures and housing allocation of residents so that phenomena under urban dynamics could be modelled. The experiments show that the proposed modelling framework could investigate the trade-off of investment on the highway and public transport system. Third, empirical observations of spatial features in transport networks are reported in this thesis. The thesis measures circuity of transport networks, because this indicator could examine how aggregate transport networks are and the efficiency of network structures. Then research methods that can deal with several data sources are developed. The empirical observation shows that there is an exponential decay between the circuity and travel time in public transport networks. Meanwhile, this thesis also presents that the average circuity in road networks is less than that in public transport networks for the same sample of trips, which to some extent show the difference of spatial features between road and public transport networks. Additionally, correlations between circuity, accessibility and mode share are analysed.