<p>Robots are expected to promote independence and improve the quality of life for older adults. Trust is a robust predictor of successful human-robot interaction, but it is rarely studied in the middle-aged and older population. This study explored the effect of ascribing mental capacities to a humanoid robot on anthropomorphism and trust. The study was based on an online survey of 411 Chinese adults aged 50 and over. Participants were primed for a robot with high versus low levels of four theoretically derived dimensions of mental capacities (values, autonomy, sociality, and emotionality) and watched a brief video depicting a humanoid robot. They then played a one-shot trust game with the robot and completed a questionnaire assessing their perceived agency and experience. Results indicate that ascribing different mental capacities to robots predicts distinct aspects of robot anthropomorphism. The mental capacities of robots had little influence on self-report trust. However, changing the emotion of the robot from low to high reduced participants’ risk-taking behavior as measured by the one-shot trust game. This negative relationship can be mitigated by perceived experience and agency. This study demonstrates that ascribing emotions to robots has a dual impact on trust behavior, supporting both the humanization hypothesis and the uncanny valley hypothesis, with important implications for robot design in aging populations.</p>