Abstract. Probabilistic model checking is an increasingly widely used formal verification technique. However, its dependence on computationally expensive numerical operations makes it particularly susceptible to the state-space explosion problem. Among other abstraction techniques, bisimulation minimisation has proven to shorten computation times significantly, but, usually, the full state space needs to be built prior to minimisation. We present a novel approach that leverages satisfiability solvers to extract the minimised system from a high-level description directly. A prototypical implementation in the framework of the probabilistic model checker PRISM provides encouraging experimental results.