This study aims to examine the validity of using non-patent reference (NPR) in a patent document to measure the performance of university-industry collaboration. A total of 77,013 patents with number of assignees equal to two are retrieved from USPTO patent database and classified into different groups based on the type of co-assignation. The fundamental assumption in literature for justifying the use of NPR to measure university-industry collaboration is that university-industry co-assigned patents should contain the highest number of NPRs per patent than other types of co-assignations, e.g. university-university or industry-industry, because university-industry co-assigned patents are inventions substantially based on the collaboration between university and industry. However, it is found in this study that university-university co-assigned patents contain the highest number of NPR (27.18 NPRs per patent) instead of the expected university-industry co-assigned patents (20.13 NPRs per patent). Also, the industry-industry co-assigned patents contain significantly low number of NPRs (2.42 NPRs per patent). This signifies that the assumption in literature is invalid and the use of NPR to measure university-industry collaboration is not proper. Furthermore, NPR can also be added by patent examiner, this enhances the uncertainty when using NPR to measure university-industry collaboration generated by patent inventor or patent applicant. According to the above reasons, this study suggests that it is not proper to use NPR as an indicator to measure university-industry collaboration.