This paper investigates the use of elitism in chemical reaction optimization (CRO) to address its convergence performance in difficult problems. We focus on problems with complex and highly discontinuous solution space. In such problems CRO’s convergence performance tend to be sluggish as the algorithm repeatedly digress from the best-found solution characteristics in search for solutions in different areas in the problem’s solution space. A complex network design problem is used to demonstrate this issue and experiment with the impact of elitism on algorithm convergence. Elitism has been used successfully in evolutionary algorithms. Results show that its use in CRO improves algorithm convergence performance drastically. However, due to CRO’s tendency to have diminishing population of molecules in such problems, the use of a larger list of elit solutions appears to be ineffective in improving the algorithm performance beyond the initial gains from introducing elitism. We investigate the reasons behind this observation and point to possible solutions.