To study the relative merits of three training principles - difficulty of training, specificity of training, and variability of training - subjects were trained to follow navigation instructions to move in a grid on a computer screen. Subjects repeated and then followed the instructions by mouse clicking on the grid. They were trained, given a short distractor task, and then tested. There were three groups, each receiving different message lengths during training: easy (short lengths), hard (long lengths), and mixed (all lengths), with all subjects given all lengths at test. At test, the mixed group was best on most lengths, the easy group was better than the hard group on short lengths, and the hard group was better than the easy group on long lengths. The results support the advantages of both specificity and variability of training but do not support the hypothesis that difficult training of the form used here would lead to overall best performance at test.