This paper presents an application where machine learning techniques are used to mine data gathered from online poker in order to explain what signifies successful play. The study focuses on short-handed small stakes Texas Hold'em, and the data set used contains 105 human players, each having played more than 500 hands. Techniques used are decision trees and G-REX, a rule extractor based on genetic programming. The overall result is that the rules induced are rather compact and have very high accuracy, thus providing good explanations of successful play. It is of course quite hard to assess the quality of the rules; i.e. if they provide something novel and non-trivial. The main picture is, however, that obtained rules are consistent with established poker theory. With this in mind, we believe that the suggested techniques will in future studies, where substantially more data is available, produce clear and accurate descriptions of what constitutes the difference between winning and losing in poker.