We present aggressive abstraction techniques for probabilistic automata (PA), a state-based model involving discrete probabilistic and nondeterministic branching. Our abstractions yield abstract PA in which transitions are typed “possible” or “required”—as in modal transition systems—and have constraint functions as target. The key idea is to focus on identifying common combined-transitions from concrete states and putting them as required ones in the abstract state. We prove the correctness of our abstraction techniques, study their relationship, and show that they are compositional w.r.t. parallel composition. We also show the preservation of probabilistic and expected reachability properties for PA.