The problem of approximation to large-scale Boolean networks is considered. First, we assume a large-scale Boolean network is aggregated into several sub-networks. Using the outputs(or inputs) of each sub-network as new state variables, a new simplified time-varying network is obtained. Then a time-invariant Boolean network is used to approximate each subsystem. Observed data are used to find the best approximating dynamic models. Finally, the aggregation method is investigated.