AbstractThis paper addresses the problem of fault tolerant estimation and the design of fault tolerant sensor networks. Fault tolerance is defined with respect to a given estimation objective, namely a given functional of the system state should remain observable when sensor failures occur. Redundant and minimal sensor sets are defined and organized into an automaton which contains all the subsets of sensors such that the estimation objective can be achieved. Three criteria, which evaluate the system fault tolerance with respect to sensor failures when a reconfiguration strategy is used, are introduced: (strong and weak) redundancy degrees (RD), sensor network reliability (R), and mean time to non‐observability (MTTNO). Sensor networks are designed by finding redundant sensor sets whose RD and/or R and/or MTTNO are larger than some specified values. A ship boiler example is developed for illustration. Copyright © 2004 John Wiley & Sons, Ltd.