In the cerebral cortex, activity dynamics measured at a mesoscopic (0.1 to 1 mm) scale are characterized by both spontaneous and task-related dynamical waves of synchronized neuronal activity ranging from spreading stationary depolarizations (blobs) to traveling waves that can follow complex trajectories throughout the cortical network. These waves are thought to participate in information propagation and processing, but remain poorly understood. To test whether cortical waves can be actively generated and their trajectory controlled, we have implemented a goal-directed task for head-fixed mice, based on the online processing of wide-field calcium imaging signals. 17 transgenic mice expressing GCaMP6f in excitatory cortical neurons (Ai-95 x EMX-Cre) were implanted with a 6 mm optical window covering the left primary somatosensory and motor cortices. Calcium-dependent optical signals were analyzed in real time, to track the trajectory of propagating mesoscale waves and condition the delivery of water rewards to a specific predefined wave trajectory. Our experiments show that mice can be trained to voluntarily control the generation of specific waves of activity in primary somatosensory and motor cortices. The mice could increase the number of waves giving a reward opportunity, as well as coordinate their licking activity with the wave. The generation of these waves correlated with stereotyped movement of the body part in correspondence to the brain area where the wave was elicited. Such operant conditioning of cortical dynamics emerging from large neuronal assemblies is a critical step to assess the potential of meso-scaled brain-machine interfaces.