This study aimed at exploring the impact of mentor text modelling on Iranian English as a Foreign Language (EFL) learners' accuracy and fluency in writing tasks with different cognitive complexity in comparison with two conventional approaches to teaching writing; namely, process-based and product-based approaches. To this end, 60 Iranian EFL learners in three comparison groups were taught English writing adopting one of the three approaches enumerated above. To investigate any changes in the learners' performance as a result of the study course, their fluency and accuracy in writing tasks with high, moderate, and low cognitive complexity levels were gauged before and after the training. Based on the results of covariate analysis, mentor text modelling yielded higher degree of fluency compared to a product-based approach; however, the approach was found to be inferior to a process-based approach in terms of enhancing fluency in writing. Additionally, Mentor text modelling was found to be as effective as a product-based approach, and at the same time, more effective than a process-based approach in improving accuracy, regardless of the writing tasks' cognitive complexity level. The findings may have several implications for EFL teachers and course designers interested in developing writing syllabuses.