In the past five to ten years, there has been a major paradigm shift in the sorts of algorithms used to play classical games of mental skill.Domains which have traditionally been resistant to the approaches that have worked so well in games like chess and checkers --- examples include go, poker and various trick-taking card games --- have now seen the appearance of serious computer contenders.  I will broadly review these trends, before discussing in greater detail the challenge of opponent modelling in imperfect information games such as poker and a German card game called Skat.  In the context of this latter game, I will highlight our work in creating one of the world's first game-playing programs that can adaptively exploit individual opponents in real time.