The classical city of Butrint in southern Albania embodies over three millennia of settlement history. A Roman colony was established sometime after 31 BC, which led to the expansion of the city southwards onto a low-lying floodplain where settlement continued well into the late antique period. In this paper we describe the results of a detailed magnetometry survey undertaken to investigate Roman settlement upon the floodplain. The study included the use of multilayer perceptron neural networks to further process the magnetic data and derive estimates of feature burial depths, allowing a three-dimensional reconstruction of buried subsurface remains to be made. The neural network approach potentially offers several advantages in terms of efficiency and flexibility over more conventional data inversion techniques. The paper demonstrates how this can lead to an enhanced interpretation of magnetic survey data, which when combined with other geoarchaeological data can provide a clearer picture of settlement evolution within the context of landscape change. The value of this processing technique is also evident within the context of cultural resource management strategies, which potentially restrict more intrusive methods of investigation.