Complementation of finite automata is a basic operation used in numerous applications. The standard way to complement a nondeterministic finite automaton (NFA) is to transform it into an equivalent deterministic finite automaton (DFA) and complement the DFA. The DFA can, however, be exponentially larger than the corresponding NFA. In this paper, we study several alternative approaches to complementation, which are based either on reverse powerset construction or on two novel constructions that exploit a commonly occurring structure of NFAs. Our experiment on a large data set shows that using a different than the classical approach can, in many cases, yield significantly smaller complements.