We exploit a controlled non-framed laboratory experiment to study settlement negotiations and the plaintiff’s decision to raise a lawsuit in case of an impasse. We find that greater variance in court outcomes increases the litigation rate. Further analysis suggests that this is due to the reflection effect in plaintiffs’ loss aversion who treat disadvantageous inequality as a loss and who are thus willing to take negative expected value bets for more equality. When studying the settlement negotiations, the best-fitting logit-quantal-response-equilibrium predicts observed comparative statics patterns not predicted by the subgame perfect equilibrium.