The construction of prediction intervals and regions and their probability content for nonlinear systems with nonparametric disturbances is considered. The semiparametric efficiency bound for estimating the probability content of a known interval (region) and estimators that attain the bound are developed. Semiparametric efficient estimation of optimal prediction intervals (regions) which either (i) maximize probability content given interval length (region area) or (ii) maximize interval length (region area) given probability content is studied. The estimated probability content of (i) is found to have the same limiting behavior as if the interval (region) were known with certainty and hence attains the semiparametric efficiency bound. Further, the estimated probability of the estimated interval (region) approximates the true coverage probability to order root-n for (i) but order smaller than root-n for (ii). A Monte Carlo experiment is conducted to compare the new predictors to competitors.