Problem-based learning (PBL) is one form of active learning for the 21st century skills development. However, using the tutorial-based assessment in PBL faces a problem of inconsistencies among students, peers and tutors in rating. This paper proposed a causal model of accuracy in tutorial-based rating in PBL, based on documentary analysis and structured interview. Participants were 5 subject matter experts in PBL of universities in Thailand. The research results were as follows: 1) The proposed model was based on literature review of 3 models of performance rating, and 17 research articles. It consisted of casual factors with two dimensions – rater context and judgment. For rater context, it was measured by five variables (goals, ability, conscientiousness, motivation, and accountability). For judgment, it was measured by two variables (standard perception and comparison process). Moreover, there was a moderator variable in this model, multi-rater groups. All of these variables may affect accuracy in tutorial-based rating in PBL. 2) Five subject matter experts opinion that the causes of the different scores among multi-rater groups (students, peers, and tutors) were lack of understanding on the assessment forms, lack of ability to use the assessment forms, biases, lack of standardization in rating and different result expectations. These causes were consistent with the variables within the proposed model. 3) More than four out of five subject matter experts agreed with the operational definitions of all factors and variables within the proposed model.