This paper considers new ways of constructing generalized empirical likelihood (GEL) specification tests for moment condition models. We explore an attractive feature of GEL estimation: the possibility of estimating a set of weights, the so-called GEL implied probabilities, which impose numerically all moment conditions on the data. The resultant GEL distribution function is a more efficient estimator of the distribution of the data than the empirical distribution function. Contrasting certain functions of these two estimators of the data generating process gives rise to Pearson-type tests of overidentifying moment conditions. We show also that a similar approach can be used to construct Pearson-type tests for parametric restrictions. A Monte Carlo study concerning tests of overidentifying moment conditions shows that the size behaviour in finite samples of one of the Pearson-type statistics proposed is clearly superior to that of alternative tests.