Introduction: Individual safety performance (behavior) critically influences safety outcomes in high-risk workplaces. Compared to the study of generic work performance on different measurements, few studies have investigated different measurements of safety performance, typically relying on employees' self-reflection of their safety behavior. This research aims to address this limitation by including worker self-reflection and other (i.e., supervisor) assessment of two worker safety performance dimensions, safety compliance and safety participation. Method: A sample of 105 workers and 17 supervisors in 17 groups in the Chinese construction industry participated in this study. Comparisons were made between worker compliance and participation in each measurement, and between workers' and supervisors' assessment of workers' compliance and participation. Multilevel modeling was adopted to test the moderating effects on the worker self-reflection and supervisor-assessment relationship by group safety climate and the work experience of supervisors. Results: Higher levels of safety compliance than participation were found for self-reflection and supervisor assessment. The discrepancy between the two measurements in each safety performance dimension was significant. The work experience of supervisors attenuated the discrepancy between self- and supervisor-assessment of compliance. Contrary to our expectations, the moderating effect of group safety climate was not supported. Conclusions: The discrepancy between worker self- and supervisor-assessment of worker safety performance, thus, suggests the importance of including alternative measurements of safety performance in addition to self-reflection. Lower levels of participation behavior in both raters suggest more research on the motivators of participatory behavior. Practical applications The discrepancy between different raters can lead to negative reactions of ratees, suggesting that managers should be aware of that difference. Assigning experienced supervisors as raters can be effective at mitigating interrater discrepancy and conflicts in the assessment of compliance behavior.