Relational Frame Theory proposes that control by novel instructions may be understood as control by networks of Same and Before or After relations. The current paper reports two experiments in which such control was demonstrated. In Experiment 1, undergraduate students were first trained to respond in accordance with Before and After relations and then trained to respond in accordance with Same and Different relations. Subjects were then presented with a number of 'instructions' in the form of networks of Same, Different, Before, and After relations in the absence of reinforcement. Of the 3 students, 2 demonstrated the required performance within two exposures to the final phase of the experiment. In Experiment 2, 5 of 8 additional subjects who demonstrated instructional control also did so in the presence of 24 novel stimulus sets without further training. The implications of these novel and generative performances for the analysis of instructional control and human language more generally are considered.