Recent analyses of Russian (Halle 1963, Lightner 1972) have been forced by the criteria of rule "naturalness" and rule "generality" to posit highly abstract underlying forms. These underlying forms and rules are claimed to represent the speaker's competence. Such analyses are now being criticized (Derwing 1973, Hooper 1974) on the following grounds: (1) highly abstract forms cannot be present in the competence of unsophisticated speakers, whose knowledge of the language must be based solely on actually heard surface forms; and (2) speakers relate alternations which are not surface-phonetically motivated, to morphological facts, rather than attribute them to a sequence of opaque phonological rules. This paper presents a new analysis of Russian conjugation which shows that whereas certain alternations once were phonologically predictable, later changes destroyed the conditions of predictability, so that underlying forms and rules were accordingly restructured. Although underlying forms in the new analysis are less abstract, and the rules less natural and general, there is reason to doubt that naturalness and generality are valid criteria for synchronic rules. Evidence from dialects is introduced to support the new analysis, which will be aimed at meeting the criticisms of earlier analyses outlined above and will contribute to a description of the competence of a speaker of Russian. (Author)