This study investigated the relative benefits for acquiring computer skills of: (1) learner-initiated versus experimenter-supplied goals and (2) active practice at selecting and applying procedures versus directed execution of procedures. Subjects with varied computer experience but no knowledge of electronic spreadsheets were randomly assigned to an interactive instruction or an exploration learning condition. Both groups read identical descriptions of 12 spreadsheet commands. The exploration learning group experimented with commands at will, setting goals, and selecting and applying procedures. The interactive instruction group did not set their own goals, but instead worked three training problems per command. The training problems for six commands were presented without solutions; subjects solved them by actively selecting and applying procedures, and afterwards received feedback. For the remaining six commands, the training problems were tutorials with explicit solutions that subjects typed in verbatim. Two days after training, all subjects solved 12 test problems. Learning commands by solving problems without explicit solutions led to longer training times, but also significantly faster and more successful performance at test than either tutorial training or exploration learning. Exploration learning did not differ significantly from tutorials in training time or performance at test. Regression analyses indicated that the advantage of problem solving was not simply due to longer training times.