This paper focuses on difficulties that primary school students have in facing mathematical word problems. In particular, we are interested in exploring how they develop in the transition from grade 2 to grade 5. The research basis of the hypothesis is that some difficulties detected in grade 5 are already predictable in grade 2. Starting from the data collected in grade 5 by the National Standardized Assessment, we carry out a quantitative analysis looking for word problems in which students experience difficulties. Subsequently, we conduct a backward analysis of the grade 2 test of the same cohort of students in order to identify a set of word problems linked with those selected in grade 5 test. The analysis shows the presence of many common difficulties in the two grades. We design and carry out specific educational activities concerning word problem-solving in grade 2. These activities produce positive changes in the experimental class compared to the control class. This could suggest that a previous intervention in grade 2 could allow overcoming future difficulties in word problem text comprehension in grade 5.