This paper examines the notion that delay in reaching agreement in bar
gaining may be caused by learning that is independent of the bargainin
g procedure. In particular, learning is not due to inference from the
observed offers and responses of the opponent, but derives from observ
ation of an exogenous, costly signal - we call this 'investigation'. F
irst we observe that even if learning is costless and perfectly inform
ative, investigation may not occur in equilibrium. Under more general
conditions, however, uninformed agents typically have an incentive to
try to manipulate their prior beliefs through investigation. The main
result is that investigation by an uninformed agent may result in sign
ificant delay occurring before agreement is reached. We show that this
delay may be sustained in the limit as the length of time period vani
shes, and that this result depends crucially on the properties of the
information production technology underlying investigation. Also, the
delay we observe is shown to be robust to changes in the bargaining pr
ocedure.