Learning consumer preferences is essential to maximize profits. To optimize the product line, accurately segmenting the market and eliciting consumer preferences in each segment are critically important. We present a robust framework to simultaneously segment the customer base and learn each segment’s preferences. We build upon ideas from machine learning and mathematical programming and propose a robust preference elicitation model. Our model accounts for robustness against feature noise (i.e., perturbations caused by consumers inaccurately comparing alternatives), and handles label noise (i.e., inconsistent consumer choices) using a weighting scheme that determines the relevance of the past choices in predicting future ones. The proposed framework has three appealing characteristics. First, it simultaneously segments the market and learns the segments’ preferences. Second, it extends an ML-based preference learning method that has been proven to be effective. Third, the decision maker can choose the level of robustness and has the option to focus on the parsimony of the solution. We perform extensive experiments and show that the proposed framework offers better prediction accuracy and lower variability in the predictions.