A fault-tolerant switching observer design methodology is proposed. The aim is to maintain a desired level of closed-loop performance under a range of sensor fault scenarios while the fault-free nominal performance is optimized. The range of considered fault scenarios is determined by a minimum number p of assumed working sensors. Thus the smaller p is, the more fault tolerant is the observer. This is then used to define a fault tolerance measure for observer design. Due to the combinatorial nature of the problem, a semidefinite relaxation procedure is proposed to deal with the large number of fault scenarios for systems that have many vulnerable sensors. The procedure results in a significant reduction in the number of constraints needed to solve the problem. Two numerical examples are presented to illustrate the effectiveness of the fault-tolerant observer design.