The fact that neural network models can approximate virtually any measurable function up to an arbitrary degree of accuracy has been emphasized. Moreover, knowledge of neuron connectivity and the choice of layer arrangement have been equally covered. With this understanding comes the daunting task of selecting the most appropriate model for a given application. This is especially true in the study of natural systems, given the unpredictable availability, quality, representativeness, and size of input data sets. Although no perfect model exists for a given application, modelers have the ability to develop and implement effective neural network techniques based on a coordinated framework such as that depicted in Figure 3.1. This circumvents many of the mistakes modelers make in their experimental approach: a line of attack with poorly stated purposes and no clear indication of model performance expectancy or validation criteria.