Abstract Lego construction ability is associated with a variety of spatial skills and mathematical outcomes. However, it is unknown whether these relations are causal. We aimed to establish the causal impact of Lego construction training on: Lego construction ability; a broad range of spatial skills; and on mathematical outcomes in 7-9-year-olds. We also aimed to identify how this causal impact differs for digital versus physical Lego construction training. One-hundred and ninety-eight children took part in a six-week training programme, delivered twice weekly as a school lunch time club. They completed either physical Lego training (N = 59), digital Lego training (N = 64), or an active control condition (crafts; N = 75). All children completed baseline and follow-up measures of spatial skills (disembedding, visuo-spatial working memory, spatial scaling, mental rotation, and performance on a spatial-numerical task, the number line task), mathematical outcomes (geometry, arithmetic, and overall mathematical skills) and Lego construction ability. Exploratory analyses revealed evidence for near transfer (Lego construction ability) and some evidence for far transfer (arithmetic) of Lego training, but overall transfer was limited. Despite this, we identified key areas for further development (explicit focus on spatial strategies, training for teachers, and embedding the programme within a mathematical context). The findings of this study can be used to inform future development of Lego construction training programmes to support mathematics learning.