The foot-and-mouth disease (FMD) virus poses a considerable threat both to farmers and to the wider economy should there be a future incursion into the UK. The most recent large-scale FMD epidemic in the UK was in 2001. Mathematical models were developed and used during this epidemic to aid decision-making about how to most effectively control and eliminate it. While the epidemic was eventually brought to a halt, it resulted in a huge loss of livestock and is estimated to have cost the UK economy around ¿6 billion. The mathematical models predicted the overall spatial spread of FMD well, but had low predictive ability for identifying precisely which farm premises became infected over the course of the epidemic. This will in part have been due to the stochastic nature of the models. However, the transmission probability between two farm premises was represented as the Euclidean distance between their point locations, which is a crude representation of FMD transmission. Additionally, the premises' point location data contain inaccuracies, sometimes identifying the farmer's residential address rather than the farm itself which may be a long way away. Local FMD transmission occurs via contaminated fomites carried by people or vehicles between premises, or by infected particles being blown by wind between proximal fields. Given that these transmission mechanisms are thought to be related to having close field boundaries, it is possible that some of the inaccuracy in model predictions is also due to imprecisely representing such transmission. In this thesis I use fine-scale geographical data of farm premises' field locations to study the contiguity of premises (where contiguous premises (CPs) are defined as having field boundaries < 15m apart). I demonstrate that the distance between two premises' point locations does not accurately represent when they are CPs. Using an area of southern Scotland containing 4767 livestock premises, I compare the predictions of model simulations using two different model formulations. The first is one of the original models based on the 2001 outbreak, and the second is a new model in which transmission probability is related to whether or not premises were contiguous. The comparison suggests that the premises that became infected during the course of the simulations were more predictable using the new model. While it cannot be concluded that this will translate into more accurate predictions until this can be validated during a future outbreak, it does suggest that the new model is more predictable in its route through the landscape, and therefore that it may better reflect local transmission routes than the original model. Networks based on contiguity of premises were constructed for the same area of southern Scotland, and showed that 90.6% (n=4318) of the premises in the area were indirectly connected to one another as part of the Giant Component (GC). The network metric of 'betweenness' was used to identify premises acting as bridges between otherwise disconnected sub-populations of premises. It was found that removing 100 premises with highest betweenness served to fragment the GC. Model simulations indicated that, even with some longer-range transmission possible, removing these premises from the network resulted in a large decrease in mean number of infected premises and outbreak duration. In real terms, premises removal from the network would mean ensuring these premises did not become infected by enhanced biosecurity and/or vaccination depending on policy. In this thesis I also considered the role of biosecurity practices in shaping FMD spread. A sample of 200 Scottish farmers were interviewed on their biosecurity practices, and their biosecurity risk quantified using a biosecurity 'risk score' developed during the 2007 FMD outbreak in Surrey. Using Moran's I and network assortativity measures it was found that there did not appear to be any clustering of biosecurity risk scores on premises. Statistical analysis found no association between biosecurity risk and the mathematical model's premises' susceptibility term (which describes the increase in a premises' susceptibility with increasing numbers of livestock). This suggests that the model's susceptibility term is not indirectly capturing a general pattern in biosecurity on different sized farm premises. Thus, this body of work shows that incorporating a more realistic representation of premises location into mathematical models, in terms of area (i.e. as fields) rather than a point, alters predictions of spatial spread. It also demonstrates that targeted control at a relatively small number of farms could effectively fragment the farming landscape, and has the potential to considerably reduce the size of an FMD outbreak. It also demonstrates that variations in premises' FMD biosecurity risks are unlikely to be indirectly affecting the spatial or demographic components of the model. This increase in understanding of how geographic, social and demographic factors relate to FMD spread through the landscape may enable more effective control of an outbreak, should there be an incursion in the UK in future.