Effects of two subtraction instruction types for low achieving children were compared. The first was based on constructivist ideas. The second was more or less traditional but did include interaction opportunities comparable to those in constructivist classrooms. Sixteen students (eight in each condition) with an average age of 10.5 years were trained during 34 lessons. Results showed no performance differences between the two groups on the subtraction problems instructed. There was a performance advantage for the traditional group on transfer problems without regrouping. The constructivist group used more different strategies but was not able to use this bigger repertory in a flexible way; instead, they used their strategies rather randomly. These findings pose questions on the usefulness of constructivist teaching in mathematics for low performers.