Traditional DEA method assumes that all inputs-outputs of observed samples are deterministic data, which restricts the practical applications of DEA. The pseudo likelihood estimation method based stochastic non-smooth envelopment of data (PLE-StoNED) in this paper extends this assumption, and can estimate frontiers in stochastic environment. The current paper proves that a frontier based on the assumptions of production possibility sets can be represented by a function with restrictions of convexity and monotonicity. Compared with the previous StoNED methods, our method can estimate the efficiencies of DMUs with multiple inputs and multiple outputs. Based on Monte Carlo experiments, the multiple inputs and multiple outputs PLE-StoNED is verified to be effective, and it can correct the bias generated by traditional methods like DEA. Finally, the new method is applied to estimate the frontier and efficiencies of commercial banks in mainland China. Our method fills up the gap of deterministic DEA method and statistical nature, which can provide decision references for decision makers who want to evaluate the productivity and efficiency for DMUs with multiple inputs and multiple outputs in stochastic environments.