Memory Based Learning (MBL) is a simple function approximation method whose roots go back at least to 1910. Training a memory based learner is an almost trivial operation: just store each data point in memory (or a database). Making a prediction about the output that will result from some input attributes based on the data is done by looking for similar points in memory, fitting a local model to those points, and then making a prediction based on the model. There are four components that define a memory based learner: a distance metric, the number of nearest neighbors, a weighting function, and a local model. Each will be described in turn in the next four subsections followed by a discussion of multivariate considerations and classification problems.