This paper describes a method eliciting componential analyses directly from monolingual native speakers. It extends Goodenough's (1951, 1956, 1964, 1965, 1967, 1968, 1970) pioneering work in the development of a method for constructing a model of how kinship categories in a society are conceptualized by its members (1964:221). Although there are alternative positions (e.g., Scheffler and Lounsbury 1971; Scheffler 1972), the approach taken here recognizes that more than one adequate structural semantic model of a single domain sometimes exists (Goodenough 1956, 1965; Wallace and Atkins 1960; Burling 1964; Wallace 1965) and that there is a consequent need to develop techniques discerning which of competing models has the greatest cognitive validity (Romney and D'Andrade 1964; Wallace 1965; Wexler and Romney 1972; Rose and Romney 1979; Romney 1980). The method of elicitation ofa componential model directly from a native speaker is one answer to Wallace's (1965) call techniques beyond the mechanics of the analysis itself constructing psychologically-real componential models. Native speaker componential models directly serve two important functions. First, the semantic dimensions isolated by the native speaker, when they coincide with those discovered by the anthropologist, can be taken as evidence of the cognitive validity of aspects of both sorts of models. It is also possible to determine whether the native speaker's insider position permits the discrimina? tion of semantic features overlooked or ignored by the anthropologist. The second major function is whole model comparison both in the assessment of the cognitive validity of competing models and in the determination of the extent of variation among native speakers. The ultimate aim of the work is to contribute to the ethnological inventory of methods and techniques permitting replicable ethnography, an essential base the development of cross-cultural (semantic) theory (Goodenough 1970, 1981). Described here are the results of a field experiment in which I sought to discover the minimum amount of instruction an intelligent Polynesian speaker required to do a componential analysis of Niutao kin terms.21 wanted to provide both an opportunity a spontaneous analysis and, should this not be forthcoming, more explicit directions accomplishing the task. Consequently, the specific method eliciting the componential analysis developed out of a more general approach in which I encouraged my co-worker to approach the organiza? tion of the kinship domain in any way she chose. The first result was a model of kin terms organized in the form of a truncated genealogical tree; only later did we