A prototype adaptive automatic control algorithm was implemented to regulate temperatures measured at several points in a tumour by adjusting the power applied to several ultrasound transducers. The goal was to control the temperatures under the elements of a mosaic applicator individually without any priori knowledge of which probes are under which elements. The control algorithm was devised for clinical applications where the position of each probe with respect to the heat sources is difficult to determine precisely. Instead, the program 'learns' the relationship between the inputs (power levels) and the outputs (temperatures) automatically. Based on the observed transfer function relating the power at m sources to the temperatures n probes, where n and m are not necessarily the same, a new method was used to implement a feedback controller. This method simplifies the design of the controller for a multiple-input/multiple-output (MIMO) system, while taking into account the coupling that may exist between the various elements of the system. As a result of using an adaptive scheme, the regulator continuously tracks changes in the system, such as blood flow variations or patient motion, by modifying its control parameters. The algorithm performance has been tested in simulations as well as experiments in dog thigh and a perfused kidney model.