This paper aims to focus on the differences between experts and users in product appearance semantic cognition and design semantic evaluation results, and explore whether a small sample of expert evaluation can replace or predict user evaluation in the preliminary evaluation of product development. A two-stage questionnaire survey was conducted. In the first stage, the design cognitive semantics of experts and users are counted, and high-frequency words are selected as evaluation semantics for next stage. In the second stage, a questionnaire survey is conducted among a smaller sample of experts and a larger sample of users. With canonical correlation analysis, the correlation results are calculated from evaluation data of experts and users, and further cluster analysis is used to check the consistency of evaluation results. Based on the carding machine appearance, 12 carding machine are selected as stimulus materials. The first sets of experts’ and users’ semantics evaluation are strongly correlated (Pearson’s r=0.938, F=2.542, p=0.054); the class-II clustering results show that the semantic evaluation results of experts and users are completely consistent, and the inconsistency rate of the class-III clustering is only 8.333%. Expert semantics and user semantics are different, reflecting the difference between experts and users' perceptions of design, but the final evaluation results are consistent. Therefore, a small sample of expert semantic evaluation can predict the results of user semantic evaluation to a certain extent, this method can help the design and development team to master and predict the design evaluation and users cognition, and save a certain amount of cost investment for textile machinery enterprises in the initial evaluation stage of design.