In this paper, stochastic production frontier models are estimated with IAB establishment data from waves 2002 and 2003 to analyze productivity and inefficiency. The data suffer from nonresponse in the most important variables (output, capital and labor) leading to the loss of 25 % of the observations and possibly imprecise estimates and invalid test statistics. Therefore the missing values are multiply imputed. The analysis of the estimation results shows that, particularly in the inefficiency submodel, working with multiply imputed data reveals some interesting and plausible results which are not available when missing observations are ignored.