Data envelopment analysis (DEA) has been widely used for efficiency evaluation of industrial plants. A conventional DEA model may easily lead to the situation where more than one-third of efficiency values are set to 1, so it is hard to analyze the pros and cons of the multi-decision-making units. The DEA cross-model can distinguish the pros and cons of the effective decision-making units, but it is unable to indicate the improvement direction of the ineffective decision-making units. This paper proposes an efficiency analysis method based on an improved DEA cross-model, which can get higher efficiency discrimination in identifying the efficiency state of the decision-making units compared with the conventional DEA model. The improved DEA cross-model avoids the impacts of unreasonable weight allocation of input and output indices. Meanwhile, self-evaluation of the improved DEA cross-model with the slack variables can find the improvement direction of the ineffective decision-making units. Also, the Malmquist productivity index (MPI) based on the improved DEA cross-model can comprehensively consider various inputoutput factors and obtain dynamic consistent analysis performance of industrial plants. To get relative effectiveness of ethylene plants under different technologies and different scales, the performance analysis of 19 ethylene plants in China were executed by proposed MPI. According to the results, various indices of MPI are evaluated over time by the improved DEA cross-model to obtain the root causes and direction of performance improvement of ethylene plants. As a result, the method proposed in this paper is effective and practical.