We develop the likelihood ratio criterion (LRC) for testing the coefficients of a structural equation in a system of simultaneous equations in econometrics. We relate the likelihood ratio criterion to the AR statistic proposed by Anderson and Rubin (1949, 1950), which has been widely known and used in econometrics over the past several decades. The method originally developed by Anderson and Rubin (1949, 1950) can be modified to the situation when there are many (or weak in some sense) instruments which may have some relevance in recent econometrics. The method of LRC can be extended to the linear functional relationships (or the errors-in-variables) model, the reduced rank regression and the cointegration models.