The recent popularity of learn-to-program websites means there are many beginner programmers learning the basics of programming with minimal human intervention. Without teachers or tutors to guide them, these beginners may get lost and frustrated in the difficult task of learning to program. In this project we automatically generate feedback taking advantage of the fact that the programs are small and we have access to large collections of correct and incorrect attempts at writing the same program. We use a combination of the TDS program synthesizer and statistics in order to go from data mining to feedback that a novice will find helpful. Our techniques are implemented as the hint mechanism on the Code Hunt programming game.