This study addresses the fundamental management problem of decision-making in a climate where future values of important variables are unknown and can at best be estimated. Traditionally various techniques have existed to help managers forecast earnings; most of these techniques are founded in statistics, thus requiring some limiting assumptions. The use of neural networks has been described as a promising non-parametric approach, negating the need for statistical assumptions. This thesis explores the application of the neural network paradigm to the area of earnings forecasting. A novel radial basis function approach is developed in several configurations and these are explored for their abilities to forecast earnings for non-financial companies included in Hong Kong''s Hang Seng 100 index. The resultant model was tested for forecast accuracy as well as its generalisability across various industries included in the index. Several incremental models are presented and tested showing that the inclusion of variables external to company information increase the accuracy of forecasts, and thus information quality.