This article addresses the problem of "education for all," and offers a research proposal that replaces procedural learning by a learning practice whereby all are engaged. Although educational research since 1990 of dialogical learning (DL) and collaborative learning (CL) have shown that it is possible to promote the learning practices that they focus on, little evidence is available on long-term effects of school achievement. Teachers also face pressure from both the UK and USA governments having produced policy documents favouring procedural teaching. An exception is CASE, Cognitive Acceleration through Science Education (1984 onwards), and a 2-year course for 12-14 year-olds. This functioned by collaborative learning placed in highly structured theory-based science lessons based on Piagetian models of difficulty. Students consistently performed higher in National exams in science, maths and English at 16 (Shayer, 1999b). It is argued that a better way of changing teaching practice would be to place it in teachers' initial training. By assisting trainee science teachers, in designing their science curriculum lessons--assisted by DL and CL literature--to extract and use the same theory-base that had been used by CA staff for constructing CASE lessons, they would possess a valid theory and practice of teaching.