Background: Methods for teaching thinking may be described as out-of-context or infusion. Both approaches have potential to raise students' general cognitive processing ability and so raise academic achievement, but each has disadvantages. Aims: To describe and evaluate a theory-based learn to think (LTT) curriculum for primary school students, which draws on the strengths of both out-of-context and infusion approaches. Sample: One-hundred and sixty-six students in three classes of Grade 1 (6+ years old), Grade 2 (7+ years old), and Grade 3 (8+ years old) in a primary school in Shanxi province, China, randomly ascribed to experimental (90) and control (76) groups. Methods: All students were pre-tested for non-verbal intelligence and academic achievement. Experimental students followed the LTT curriculum (one activity every 2 weeks) for 4 school years. All were post-tested on three occasions for thinking ability and four times for academic achievement. Results: Grade 1 and Grade 2 students showed effects of LTT from 1 year after their start and increasing: on thinking ability d = 0.78-1.45; on Chinese d = 0.68-1.07; on maths 0.58-0.87. Grade 3 students showed effects from 6 months after their start: on thinking ability 0.90-1.37; Chinese 0.77-1.32; maths 0.65-1.29. The effects were concentrated in students in the middle band of initial ability. Conclusions: A curriculum for teaching thinking based on a structured theoretical model that combines elements of out-of-context and infusion methods has been shown to have long-term far transfer effects on students' thinking ability and academic achievement. More work is needed to meet the needs of a wider range of abilities.