Data envelopment analysis (DEA) is extended to the case of stochastic inputs and outputs through the use of chance-constrained programming. The chance-constrained envelope envelops a given set of observations "most of the time." We show that the chance-constrained enveloping process leads to the definition of a conventional (certainty-equivalent) efficiency ratio (a ratio between weighted outputs and weighted inputs). Furthermore, extending the concept of Pareto and Koopmans efficiency to the case of chance-constrained dominance (to be defined), we establish the identity of the following two chance-constrained efficiency concepts: (i) the chance constrained DEA efficiency measure of a particular output-input point is unity, and all chance-constraints are binding; (ii) the point is efficient in the sense Pareto and Koopmans. Finally we discuss the implications of our approach for econometric frontier analysis.