Almost every form of statistical and machine learning method has been applied to learning QSARs at one time or another: linear regression, decision trees, neural networks, nearest-neighbour methods, support vector machines, Bayesian networks, relational learning, etc. These methods differ mainly in their a priori assumptions they make about the learning task. Most of these methods assume propositional data in tables of tuples of attributes, giving compatibility with standard chemoinformatic representations. Learning methods generally also have options and parameters that need to be selected to fully specify them, for example the type of kernel used, or the number of neighbours used in k-nearest-neighbour. This means that we will need to investigate the space of learning methods X associated parameters.