Composite likelihood functions are convenient surrogates for the ordinary likelihood, when the latter is too difficult or even impractical to compute, and they may be more robust to model misspecication. One drawback of composite likelihood methods is that the composite likelihood analogue of the likelihood ratio statistic does not have the standard 2 asymptotic distribution. Invoking the theory of unbiased estimating equations, this paper proposes and discusses the computation of the empirical likelihood function from the unbiased composite scores. Two Monte Carlo studies are performed in order to assess the nite-sample performance of the proposed empirical composite likelihood procedures.