Estimation of swell conditions in coastal regions is important for a variety of public, government, and research applications. Driving a model of the near-shore wave transformation from an offshore global swell model such as NOAAWaveWatch3 is an economical means to arrive at swell size estimates at particular locations of interest. Recently, some work (e.g. Browne et al. [Browne, M., Strauss, D., Castelle, B., Blumenstein, M., Tomlinson, R., 2006. Local swell estimation and prediction from a global wind-wave model. IEEE Geoscience and Remote Sensing Letters 3 (4), 462–466.]) has examined an artificial neural network (ANN) based, empirical approach to wave estimation. Here, we provide a comprehensive evaluation of two data driven approaches to estimating waves near-shore (linear and ANN), and also contrast these with a more traditional spectral wave simulation model (SWAN). Performance was assessed on data gathered from a total of 17 near-shore locations, with heterogenous geography and bathymetry, around the continent of Australia over a 7 month period. It was found that the ANNs out-performed SWAN and the non-linear architecture consistently out-performed the linear method. Variability in performance and differential performance with regard to geographical location could largely be explained in terms of the underlying complexity of the local wave transformation.