In simultaneous equation models, a necessary and sufficient condition for identification. One way of stating the rank condition is as follows: for each equation in the model, the matrix of parameters from all equations associated with all the variables excluded from the equation in question must have rank equal to one less than the number of equations in the system. In practice, as you might have surmised, the rank condition is difficult to check. If your model meets the order condition, and you have successfully obtained reasonable-looking parameter estimates from 2SLS, then almost surely your model is identified.
