In this project a method of pattern completion based on networks of artificial neurons (neural networks) will be employed. The artificial neural network, in effect, extracts the statistics of the spatially distributed field from punctual observations and enables the field to be estimated at unobserved locations. The method has some characteristics similar to kriging, but evolves from a completely different perspective,. It also bears many striking similarities to discriminant analysis. This similarity is exploited to provide a method of generating random field samples for Monte Carlo simulation that are conditioned on information,. The methods can be implemented very efficiently on massively parallel computers and are amenable, in principle, to implementation in computer hardware.