This paper investigates possible improvements of abstraction to simplify finite-state machines during compositional nonblocking verification of large discrete event systems. Current methods to simplify finite-state machines depend on the absence of transitions from the states to be simplified, and selfloop transitions, i.e., transitions with the same source and target state, are a common culprit that prevents simplification. Some methods to remove such selfloops are known, but they require events that appear on selfloops in the entire finitestate machine to be simplified. The methods described in this paper improve on this, because they allow for the removal of individual selfloop transitions from a finite-state machine while preserving conflict equivalence. This makes it possible to remove more transitions, thus reducing the computational effort of compositional nonblocking verification. Two abstraction rules are proposed, and experimental results show the potential of improvement over previously used methods.