The present study analysed the teaching-learning strategies taking place in classrooms where an educational innovation was implemented. This study is a sequel to a broader, quantitative year-long study by Rojas-Drummond et al. (2016). The latter, original study, included 120 sixthgraders from two (experimental and control) state schools. Both groups solved an individual and group pre- and post-intervention Test of Textual Production. Between tests, children from the experimental group participated in a program called 'Learning Together', which promoted scaffolding, collaborative problem-solving and dialogic interactions. Five lessons from experimental and control groups were video-recorded throughout the year. Macro-analyses carried out as part of the original study revealed that the experimental group (in comparison with the control), learned to compose higher quality written articles when working in small-groups and independently. In the present study, we carried out fine-grained analyses of selected lessons from each group to understand how the achievements of 'Learning Together' children might have come about. Results showed that the teaching-learning strategies of the control group tended to be directive and transmissional and literacy was addressed in a somewhat fragmented and decontextualised way. In contrast, the 'Learning Together' classroom practices showed an orientation towards the joint construction of knowledge among teachers and students through dialogic interactions and co-regulatory processes, as well as the promotion of literacy as an articulated, situated social practice. We hypothesise that these contrasting teaching-learning styles partly account for why 'Learning Together' participants became more expert writers by the end of the academic year, in contrast with their control peers.