In this paper, fault-tolerant control is viewed as the control of a finite state stochastic failure process. Coverage of failures serves to effect the control action which attempts to minimize the risk of the system level failure. A Markov model is used to describe the failure process. Transient events are scrutinized that occur between the onset of a subsystem failure and the settlement of the continuous state trajectory of the controlled system at a new equilibrium. The relationship of the coverage to the quality of diagnosis and control is analytically expressed, and the monotonic dependence of the system level reliability on the coverage is formally established. A flight control example is presented to explain the procedures for reliability analysis, coverage calculation, the execution of supervisory control, and the construction of fitness landscapes that define the place of a discrete state in the continuous state control system. (C) 2003 Elsevier Ltd. All rights reserved.