A review of the econometric literature on instrumental variables (IV) estimation shows that the performance of traditional IV estimation relies critically on the quality of the instruments. We discuss three different approaches that do not require the availability of observed instrumental variables: the ‘Higher Moments’ (HM) estimator, the ‘Identification trough Heteroscedasticity’ (IH) estimator, and the ‘Latent Instrumental Variable’ (LIV) approach. These methods attempt to identify the regression parameters not through observed instruments but by using other information that enables identifiability. The performance of these methods is illustrated on simulated and empirical data. Copyright © 2009 John Wiley & Sons, Ltd.