A well known problem in diagnosis is the difficulty of providing correct diagnostic conclusions in light incorrect or missing data. Traditional approaches to solving this problem, as typified in the domains of various complex mechanical systems, validate data by using various kinds of redundancy in sensor hardware. While such techniques are useful, we propose that another level of redundancy exists beyond the hardware level, the redundancy provided by expectations derived during diagnosis. That is, in the process of exploring the space of possible malfunctions, initial data and intermediate conclusions set up expectations of the characteristics of the final answer. These expectations then provide a basis for judging the validity of the derived answer. We will show how such expectation- based data validation is a natural part of diagnosis as performed by hierarchical classification expert systems.