This paper develops a formalism for the concept of context dependent similarity. The "context" is a diverse reagent, for example a serum containing many different antibodies. From the point of view of a serum A, serum B and serum C may be twice as similar as (say) serum B and serum D, while from the point of view of a serum E the converse may be true. Methods for measuring similarity coefficients are outlined, together with possible applications to diagnostic and prognostic tests. Individuals with a given disease or who are predisposed to develop a given disease are likely to have sera that are similar from the point of view of an appropriately chosen test serum. Possible applications to lupus and AIDS are discussed. Some of the work in the currently hot field of proteomics is similar in spirit to the ideas presented here. (The paper does not include the word proteomics, since that word more is a recent invention.)