This study empirically quantifies the efficiency of a real-world bargaining game with two-sided incomplete information. Myerson and Satterthwaite (1983) and Williams (1987) derived the theoretical ex-ante efficient frontier for bilateral trade under two-sided uncertainty and demonstrated that it falls short of ex-post efficiency, but little is known about how well bargaining performs in practice. Using about 265,000 sequences of a game of alternating-offer bargaining following an ascending auction in the wholesale used-car industry, this study estimates (or bounds) distributions of buyer and seller values and evaluates where realized bargaining outcomes lie relative to efficient outcomes. Results demonstrate that the ex-ante and ex-post efficient outcomes are close to one another, but that the real bargaining falls short of both, suggesting that the bargaining is indeed inefficient but that this inefficiency is not solely due to the information constraints highlighted in Myerson and Satterthwaite (1983). Quantitatively, findings indicate that over one half of failed negotiations are cases where gains from trade exist, leading an efficiency loss of 12-23% of the available gains from trade.