"Imagine that you have apples and bananas. Your apples and bananas are in one pile, but you want them to be in separate piles. You don't want to pay some worker to separate the fruits, so you're building a machine to do it. This machine only has two pieces of information about them: their size and how yellow they are. What it needs is a function to discriminate between apples and bananas, so that it can sort them.
One way to find a function to discriminate between two classes of things is to use a perceptron. A perceptron is a simple iterative algorithm for finding a discriminant function; in other words, it can find a function to separate our apples and bananas. Although they can only separate two classes of data, and have some other limitations, perceptrons are still an interesting introduction to learning techniques."
kuro5hin.org - Perceptrons: Intro to Machine Learning