This paper describes a computational approach to certain problems of anaphora in natural language and argues in favor of formal meaning representation languages (MRLs) for natural language. After presenting arguments in favor of formal meaning representation languages, appropriate MRLs are discussed. Minimal requirements include provisions for constructions equivalent to lambda expressions and for the separation of descriptional and assertional information types. A series of examples is discussed, in order to insure that, (1) in cases where the antecedent of an anaphoric expression is not explicitly present, it can often be derived through purely syntactic manipulations of an appropriately structured MRL and (2) the properties proposed for a logical MRL make such manipulations simple to express and to apply. Given an appropriate representation language, it is possible to deal with a rich class of anaphora. (AA)