We consider two stage least squares (2SLS) estimators of a simple simultaneous equations model when identification fails asymptotically. We investigate how the limiting distribution of the estimator changes as we vary our parametrization to allow for increasing degrees of non-identification.