We show that the pseudo empirical maximum likelihood estimator can be recast as a calibration estimator. The process of estimating the probabilities pk of the distribution function can be done also in a maximum entropy framework. We suggest that a minimum cross-entropy estimator has attractive theoretical properties. A Monte Carlo simulation suggests that this estimator outperforms the PEMLE and the Horvitz-Thompson estimator. This is a joint SALDRU/DataFirst Working Paper as part of the Mellon Data Quality Project. For more information about the project visit www.datafirst.uct.ac.za.