Traditionally, in sports, we spend the majority of practice time trying to do the "right" things; that is, producing the desired performance outcomes (e.g., hitting a golf shot down the middle of the fairway, shooting a basketball into the net, hitting a baseball into fair play). An alternative is to practice doing the "wrong" thing (i.e., an undesired outcome) sometimes. This could give the performer more information about the solution space and available affordances. In two experiments, a baseball batting virtual environment was used to compare these two-types of training. In Experiment 1, the constant goal group always attempted to "hit the ball hard into fair play" (i.e., the desired outcome) and were given technical instructions about the swing and corrective feedback from an experienced coach. The variable goal group received no technical instruction and had different outcome goals for each session, including undesired ones: Hit the ball as far to the right (or left) as you can, pop the ball up into air, and drive the ball into the ground. In Experiment 2, the constant goal group did not receive explicit instructions or corrective feedback. All groups trained for 6 weeks. In pre- and posttests, batters completed simulated games in which the pitch speed, type, and location were varied, an umpire called balls and strikes, and the fielders shifted locations (e.g., they all moved to the left or right side of the field). In both experiments, the variable goal group had a significantly larger increase in hits, adapted more effectively to variations in pitch trajectory, and demonstrated superior affordance perception (e.g., not swinging at pitches out of the strike zone and ones that would have likely led to hitting into the defense).