Motivated by the savings in communication bandwidth and sensor transmission energy, data selection for estimation with wireless sensor networks is investigated in this paper. Existing approaches to data selection inherently treat sensing and transmission to a central fusion unit as of equal cost. However, energy expenditure in sensing is generally a fraction of that needed for communication. To alleviate the latter, measurement censoring at sensor nodes is proposed here for data reduction, along with a novel maximum likelihood estimator that optimally incorporates knowledge of the censored data model. Furthermore, a closed-form expression for the Cramér-Rao lower bound on the estimator variance is presented. Numerical studies show that the estimator using censored measurements achieves error values that are competitive with alternative methods, under various sensing conditions, while retaining lower computational complexity.