We use a laboratory experiment to study bargaining with random implementation. We modify the standard Nash demand game so that incompatible demands do not necessarily lead to the disagreement outcome. Rather, with exogenous probability q, one bargainer receives his/her demand, with the other getting the remainder. We use an asymmetric bargaining set (favouring one bargainer) and disagreement payoffs of zero, and we vary q over several values. Our results mostly support game theory s directional predictions. As with conventional arbitration, we observe a strong chilling effect on bargaining for q near one: extreme demands and low agreement rates. Increasing q reinforces the game s built-in asymmetry - giving the favoured player an increasingly large share of payoffs - but also raising efficiency. These effects are non-uniform: over sizable ranges, increases in q have minimal effect, but for some q, small additional increases lead to sharp changes in results.