This is an interim report on a project to construct an agent-based simulation that reproduces some of the interactions between students and their teacher in classroom lessons. In a pilot study, the activities of 67 students and 7 teachers during 40 lessons were recorded using a data collection instrument that currently captures 17 student states and 15 teacher states. These data enabled various conceptual models to be explored, providing empirical values and distributions for the model parameters. Using these data, a lesson can be ‘played back’ using a visualization program implemented in NetLogo. A visualization and simulation can be viewed side-by-side and their outputs compared in various ways, e.g. overall class state-transition matrices or individual student state trajectories. The main challenges are the formulation of descriptive rules, establishing what metrics to use to compare lessons, and determining how to validate a simulation.