AbstractThere is a type of analogical change called ending transfer, the analogical extension of an ending from one noun declension to another within a language. There are two types of ending transfer: one is the transfer of a case ending only, the other is the transfer of an apparent stem formative along with the case ending. Evidence of various kinds suggests that both types are actually transfers of case endings only. Thus, ending transfer evidence reveals whether or not speakers have posited a stem formative morpheme in certain declensions. Inferences can be drawn from this information about what kinds of language data determine speakers' morphological analyses. The findings are that speakers are influenced by declension‐internal data only, namely, the degree to which phonetic material following the root recurs in the case/number forms of the declension, and the number of alternants a hypothetical stem formative has within the declension. Declension external data such as case endings from other declensions and facts about word formation play no role. These findings provide additional support for a hypothesis drawn from other research that one factor influencing morpheme abstraction is recurrence of phonetic material in similar words, whether or not such material has a clearly assignable meaning.In this paper, some evidence will be presented from language change data on how speakers abstract and segment stem formatives and case endings in noun declensions, and on what aspects of language data such analyses are based. This study is intended both to contribute to our knowledge about psychologically real morphological analyses of inflected nouns, and to furnish evidence that is useful in developing a theory of psychologically real morphology.The first section consists of a presentation of the possibilities which speakers theoretically have for segmenting noun declensions into morphemes. The following section is a set of arguments that a certain type of analogical change which I call ending transfer provides evidence on how speakers have segmented inflected forms of nouns. The final two sections of this paper present conclusions about what aspects of language data contribute to speakers' morphological analyses of noun declensions, and relate these conclusions to other research on morphology, to the theory of generative phonology, and to a new theory of morphology.