Experimental studies of two-person sequential bargaining demonstrate that the concept of subgame perfection is not a reliable point predictor of actual behavior. Alternative explanations argue that 1) fairness influences outcomes and 2) that bargainer expectations matter and are likely not to be coordinated at the outset. This paper examines the process by which bargainers in two-person dyads coordinate their expectations on a bargaining convention and how this convention is supported by the seemingly empty threat of rejecting positive but small subgame perfect offers. To organize the data from this experiment, we develop a Markov model of adaptive expectations and bounded rationality. The model predicts actual behavior quite closely.