Outcomes of stated choice surveys are widely used for estimating choice models. Generating good experimental designs for these surveys is an important task. Using efficient choice designs typically have the advantage over orthogonal choice designs that they require fewer respondents for the same reliability of parameter estimates. However, prior parameter values are needed for generating efficient designs and efficiency may be lost if the priors are incorrect. In this paper we propose what we call serial choice conjoint in which each respondent faces a survey that has been optimized based on the choices of previous respondents. A new efficient design is generated for each respondent based on parameter estimates from choice data of all previous respondents. Therefore, there is no need for prior parameter values as input, as these will be automatically continuously updated. This concept has been proposed by others (e.g. Kanninen, 2002) but never implemented or analyzed in more detail. In contrast to adaptive conjoint, the choice situations in serial conjoint experiments do not change within-respondent, therefore estimation biases are not expected. The principle of these serial efficient designs keeps the advantage of efficient designs, i.e. requiring less respondents, while avoiding the need of priors. Preliminary results using synthetic sampled respondents show that starting off with an orthogonal design (requiring no priors) in a serial choice conjoint setting requires only slightly more respondents than starting off with an optimal efficient design based on correct priors (which are typically unknown) and requires much less respondents than just keeping the fixed orthogonal design. The disadvantage of serial choice conjoint is that respondents cannot answer the questions in the survey in parallel, but only in a serial manner. Surveys with CAPI (computer-aided personal interviewing) are therefore still possible; however internet surveys will be less suitable or need smart implementations.