This study aims to identify errors made by primary school students when modelling word problems and to eliminate those errors through scaffolding. A 10-question problem-solving achievement test was used in the research. The qualitative and quantitative designs were utilized together. The study group of the quantitative design comprises 248 elementary 4th grade students attending nine classes at three state schools in the city centre of Kütahya, chosen with the cluster sampling method. Frequency analysis and discriminant analysis were performed to analyse the quantitative data. The qualitative data were collected through clinical interviewing. The study group with whom the clinical interviews were performed comprises 30 primary school students in the class closest to the average problem-solving achievement among the nine classes. As a result, it was observed that most of the errors made by the students were caused by the use of the number operator model, which was followed by incorrect relations, number consideration, missing critical information, an inability to determine structure and relation and incorrect diagrams. The discriminant analysis shows that the biggest contribution to discriminating between students with high and low levels of modelling achievement is made by errors originating from using the number operator model, and this type of error is followed by incorrect relations, an inability to determine structure and relation and number consideration models respectively. It was concluded that errors originating from missing critical information are mostly made by successful students and the ratio of errors originating from incorrect diagrams does not affect the distinction between successful and unsuccessful students. The research also found that the modelling cycle of students does not benefit from the interpretation and validation stages. Finally, it was seen that more than half of errors made during modelling can be corrected through scaffolding.