During normal operations, response or quality variables of a process will follow a input-output relation that depend on certain key sensor variables. A soft-sensor with limited model size can be developed. When the process enters into a “faulty” state, the structure, of this model may change. Moreover, disturbance that is not reflected by sensor outputs may also be present. In this paper, we introduce a recursive soft-sensor modeling algorithm which employs a nonnegative garrote (NNG) variable selection procedure. This model can be used for both prediction, and detection of structural model change and the emergence of disturbance. The advantages of the proposed method were demonstrated by a simulation example and an industrial application to temperature prediction of a blast furnace hearth.